Retire the published semantic-pack cases

[DavidBakerEffendi]

Closes #180.

Both cases have errored on every run since the v0.9.1 bundle stopped being
reachable, and errors == 0 is a freeze gate, so the development freeze path
could not produce a clean run.

Republishing was tried first, and got one gate further

The artifact is recoverable byte-for-byte. It is now archived on the mirror
under the asset-only tag semantic-packs-v0.9.1, serving exactly the
a01cb3d7… the cases pinned, and download plus checksum verification both
pass. The next gate does not:

verify semantic-pack release bundle: unsupported release bundle schema 1

v0.10.8 requires release-bundle schema 2 on exact equality — !=, not a
minimum — and checks the generator identity immediately afterwards. These packs
cannot be consumed by any current pin regardless of where they are hosted, so
hosting was never the real blocker.

Keeping the cases would have meant re-authoring reviewed ground truth against a
schema-2 bundle: new bundle, artifact and semantic hashes for each, plus fresh
content-addressed bifrost-model:// expectation URIs derived from them.

The published-pack surface was a casualty of the open-core/private migration
rather than a contract the corpus still needs, so the two documents and their
fixtures are retired.

Deliberately kept

• Harness support. The semanticPacks schema, document model and runner
  path are untouched, so a schema-2 pack case can be authored later without
  rebuilding any of it.
• The archived bundle, left published on the mirror. It costs nothing and
  restores a content-addressed artifact any external consumer pinning that
  checksum can resolve again.

Effects

	before	after
documents	61	59
authored cases	232	230
planned cases	226	224

No promotion or cohort artifact moves: the legacy cohort already skipped these
documents as postdating the frozen 158-case inventory, and the promotion
manifest never bound them.

requested_totals_cover_the_full_authored_corpus caught the corpus change and
is updated. The evidence map drops its semantic-pack counts, and its generator
now omits the clause entirely rather than publishing "0 semantic-pack cases".

Verification

• cargo test --locked — 250 passed, 0 failed
• python3 -m unittest discover tests — 42 tests, OK
• validate benchmarks/cases — 59 files
• cohort, promotion (v1 and v2), and reproduction contract all clean
• generate-docs-evidence.py --check clean

🤖 Generated with Claude Code