Pack the wasm archive somewhere a consumer can link it

[jcant0n]

The package has been shipping 3.6 MB of browser-wasm that nobody could have used. Two independent mistakes had to line up:

• The archive was packed as libJoltC.a while every DllImport says "JoltC". On wasm the module name comes from the file name, so the lib prefix alone breaks resolution.
• No buildTransitive targets file, which matters more. runtimes/browser-wasm/native is never probed — wasm has no dynamic loading, so the archive has to be linked in at publish time via NativeFileReference. Without one it was not linked badly, it was not linked at all.

Nothing could have caught it. The eleven C suites in JoltPhysicsC test the library, not the package. The CD proves the package builds. The desktop legs load out of runtimes/ at run time, which is precisely the mechanism wasm does not use. It surfaced by opening the published .nupkg to check something else.

The archive was fine all along

Before changing anything I renamed the published .a to JoltC.a and linked it into a real .NET wasm application. It worked:

JoltC on browser-wasm
  allocator=0x7e7300 jobSystem=0x7e6dc0
  sphere went from y=2.000 to y=0.486 in 63 steps
PASS

None of what made Cesium's wasm hard is present — no setjmp in any form, no pthread, no JS exception model — so no emcc flags change in JoltPhysicsC, and the contingency about porting JobSystemSingleThreaded is moot: numThreads = 0 is enough.

And the leg that would have caught it

WasmSmokeTest consumes the package through PackageReference, so the targets file and the packed file name are part of what is under test. A ProjectReference bypasses both and would pass against the package this PR fixes.

It is its own job rather than a matrix row: no identifier to select, a workload to install, and node rather than the host to run the result. Verified locally against a package packed from this branch:

  ok    the allocator and job system were created, so the archive linked and the DllImports resolved
  ok    the physics system was created
  sphere went from y=2.000 to y=0.486 in 63 steps
  ok    the sphere came to rest in 63 steps, so the simulation both ran and settled
  ok    the sphere fell, so gravity was integrated
  ok    it settled on the floor rather than falling through it (y=0.486)
PASS

One difference from Cesium worth noting: there the ten natives are fetched at CD time, so its equivalent block is conditioned on the file existing. Here they are committed, so a missing archive means a bad fetch and packing should fail loudly instead — the condition is deliberately absent.

[github-actions]

API gate: additive

Every symbol that existed still exists, unchanged. Nothing that compiled before stops compiling.

Target framework	Added	Removed
net10.0	+0	-0

_{Enum and constant values are part of the measured surface: a renumbering keeps compiling and sends the wrong number to the driver, so it counts as a removal.}

[github-actions]

API gate: additive

Every symbol that existed still exists, unchanged. Nothing that compiled before stops compiling.

Target framework	Added	Removed
net10.0	+0	-0

_{Enum and constant values are part of the measured surface: a renumbering keeps compiling and sends the wrong number to the driver, so it counts as a removal.}