Fix the dead endpoints, dead paths and dead files the samples audit found

[romanett]

Proposed changes

An audit of the samples against the feature documentation of the stack
(UA-.NETStandard/docs, at the 2.0.0-preview.4 version
targets.props pins) turned up three kinds of
problem. This PR fixes all of them. The coverage gaps the same audit found are
filed separately as #862 and are not in scope here.

1. Every server sample advertised an endpoint it never served

All 21 server configurations declared an https base address next to their
opc.tcp one, and no sample anywhere called AddHttpsTransport(). The
stack creates a listener only for a scheme whose factory is registered and
skips the rest without an error, while GetDiscoveryUrls() keeps advertising
the address — so every one of these servers published a discovery URL that
refuses connections.

Fixed by capability rather than uniformly:

• ReferenceServer, Server.Net4 and Client.Net4 already reference
  Opc.Ua.Bindings.Https, so they now call AddHttpsTransport() and keep
  their endpoint.
• The 17 quickstart servers and the 2 GDS servers do not reference it, so they
  drop the address rather than gaining an ASP.NET Core dependency.
• SampleApplicationHostedService now hands the server the container's
  ITransportBindingRegistry. Without this, a registration made on the
  ApplicationInstance path would never reach the server it was made for,
  because ServerBase falls back to a private registry that only knows
  opc.tcp.

Why this was invisible: the tier 2 harness (SampleServerHost.KeepOpcTcpEndpointsOnly)
strips every non-opc.tcp base address before starting a server, so no test
ever exercised the configurations as shipped. A new tier 0 test,
BaseAddressesOnlyUseRegisteredTransports, closes that hole: a config may
declare a base address of another scheme only if its sample is on an explicit
allowlist of samples that register one. Verified it fails when the condition is
reintroduced (1 failure / 43 passes), then reverted the probe.

2. Dead paths a user could walk into

The UserAuthentication server advertised an IssuedToken policy for the
WS-Security Kerberos profile with an IssuerEndpointUrl naming a machine that
has not existed for years, and the Kerberos tab of its client threw
NotSupportedException because the token provider it needs does not exist on
modern .NET. Both removed, with a pointer to docs/IdentityProviders.md for the
flows the stack does support. The unused SAML helper beside it goes too.

Views client connected with .Wait() on the UI thread; now async void like
every sibling sample.

3. SDK functionality the samples reimplemented or ignored

• TimeProvider went from zero uses to threaded through every simulation.
  38 files read DateTime.UtcNow and ten created raw Timers, so a server run
  against a fake clock kept simulating on the wall clock. The node managers now
  resolve the server's clock through the documented ITimeProviderProvider
  seam and create their timers from it. PerfTest measures its update duration
  with GetElapsedTime instead of subtracting two wall-clock readings.
• The NodeManagement client replaces a hand-written EventFilter and
  publish pump with the stack's ModelChangeTracker, which also evicts changed
  nodes from the INodeCache — the part the hand-written version did not do,
  so the client could read from a stale cache after a model change.
• ReverseConnectManager is disposed asynchronously from StopAsync instead of
  through the obsolete synchronous Dispose; ArgumentListCtrl awaits
  GetBuiltInTypeAsync.

The solution now builds with zero CS0618, down from 10.

4. 50 dead files deleted

Everything under the item groups labelled "Compile items now included by
globbing that were not in the original project file" — SDK-style migration
leftovers that nothing has compiled since, several of them calling APIs 2.0
removed. 13 dialogs/controls in ClientControls.Net4 (5 superseded by the live
copies under Common/Client), the whole ReferenceServer/Aggregates folder
(predates the stack's AggregateManager; verified neither consumer references
that project), and three superseded type files in the AlarmCondition and
HistoricalEvents clients.

The two remaining Compile Remove entries are conditional TFM excludes, not
dead code, and stay.

5. README drift

The console walkthrough pointed at Samples/NetCoreConsoleServer and
Samples/NetCoreConsoleClient, neither of which exists. Rewritten around
the console samples that do, with the target frameworks (net10/net8/net48, not
".NET Core 2.0 and UWP"), solution file (UA Samples.slnx) and IDE version
corrected, ConsoleLds added to the project table, and a pointer to the
samples that live in the stack repository.

Related Issues

• Coverage gaps from the same audit, tracked separately: Samples audit: SDK functionality with no sample coverage #862
• Two upstream defects this work surfaced:
  • Base address with an unregistered transport is skipped silently, but still advertised UA-.NETStandard#4416 — the silent skip in §1 is a stack
    behaviour, and docs/Transports.md documents the opposite ("startup throws
    a clear error").
  • NodeBrowser has no async seam, so a browse that needs I/O must block UA-.NETStandard#4415 — NodeBrowser has no async seam, which
    is why Workshop/Aggregation/Server/Browser.cs still blocks on a task. It
    cannot be fixed in this repository; the constraint is now documented in the
    code instead of left looking like sloppiness.

Types of changes

• Bugfix (non-breaking change which fixes an issue)
• Enhancement (non-breaking change which adds functionality)
• Test enhancement (non-breaking change to increase test coverage)
• Breaking change (fix or feature that would cause existing functionality to not work as expected, requires version increase of Nuget packages)
• Documentation Update (if none of the other choices apply)

Checklist

• I have read the CONTRIBUTING doc.
• I have signed the CLA.
• I ran tests locally with my changes, all passed.
• I fixed all failing tests in the CI pipelines.
• I fixed all introduced issues with CodeQL and LGTM.
• I have added tests that prove my fix is effective or that my feature works and increased code coverage.
• I have added necessary documentation (if appropriate).
• Any dependent changes have been merged and published in downstream modules.

All four tiers pass locally, unchanged before and after the deletions:

Tier	Tests	Result
0 — Configuration	227	pass
1 — Server smoke	115	pass
1.5 — Node managers	165 (3 skipped)	pass
2 — Client smoke	61	pass

Further comments

Three findings from the audit did not survive checking, and are worth stating
so nobody re-reports them. The samples do not have a stale security-policy
problem: the empty <SecurityPolicyUri></SecurityPolicyUri> entries are a
wildcard that ValidateSecurityPolicies expands to Basic256Sha256 +
Aes128_Sha256_RsaOaep + Aes256_Sha256_RsaPss, and the Basic256 entries are
inside a <!-- deprecated --> block. Only ECC is genuinely missing, and that
needs an ECC application certificate rather than a config line, so it moved to
#862.

On scope. The endpoint fix could have gone the other way — add
Bindings.Https to all 21 servers. I chose not to: a quickstart that exists to
demonstrate one Part of the spec should not gain an ASP.NET Core dependency to
serve an endpoint nobody uses. The three samples that already carry the
dependency keep their endpoint and now actually serve it.

On ITimeProviderProvider over constructor injection. The migration doc
suggests a nullable TimeProvider as the last constructor parameter. That does
not work uniformly here, because nine of the sample node managers are
source-generated and their constructors are emitted. The ITimeProviderProvider
seam works identically for generated and hand-written managers and resolves the
server's clock rather than a separately injected one — which matters, because
SampleServerFactory deliberately drops the TimeProvider the stack hands it.

🤖 Generated with Claude Code