Stop conjoined_multi_tenancy and its partitioning twin sharing a schema

[jeremydmiller]

Fixes the CI failure that blocked #5085 from merging. Not caused by that PR — it is a latent isolation bug of the same family as #5070, exposed by xunit v3's parallelism rather than created by it.

The failure

await session.Query<User>().CountAsync()  should be 2 but was 4

hitting can_query_on_multi_tenanted_and_non_tenanted_documents in both conjoined_multi_tenancy and conjoined_multi_tenancy_with_partitioning, in the same run. Two independent flakes don't synchronise like that; a contended shared resource does.

Cause

Both classes build their own DocumentStore on the same hard-coded schema:

opts.DatabaseSchemaName = "mixed_multi_tenants";

then DeleteAllDocumentsAsync(), bulk-insert two User rows, and assert the count is 2. Neither class carried a [Collection] attribute, so xunit v3 puts them in separate collections and runs them in parallel — both insert before either counts, and both see 4.

Fix, from both directions

• The two classes now share the mixed_multi_tenancy collection, so xunit will not run them concurrently.
• conjoined_multi_tenancy gets its own schema, mixed_multi_tenants_conjoined, so they no longer touch the same tables at all.

The docs wrinkle

The schema name sits inside #region sample_tenancy-mixed-tenancy-non-tenancy-sample in both files — and mdsnippets renders both into docs/documents/multi-tenancy.md, the second as a duplicate -sample-1 anchor. (That duplication is a pre-existing wart worth cleaning up separately: the same sample is published twice.)

So the rename was applied to the copy that only feeds the duplicate anchor. The primary sample readers actually see is byte-identical; the regenerated duplicate block is committed. The [Collection] attributes sit outside the region markers, so they don't appear in the docs at all.

Verification

• DocumentDbTests: 1089/1090, 0 failures on both net9.0 and net10.0.
• mdsnippets gate clean — the only doc content change is the one schema line inside the duplicate block, confirmed by line position to be after the -sample-1 anchor and not in the primary. markdownlint and cspell clean.

Being straight about what the green run does and doesn't prove: the suite also passed on clean master, so a passing local run does not by itself demonstrate the fix — the collision needed CI's parallel timing to surface. The justification is the mechanism (shared literal schema + separate collections + wipe-then-count), and CI is where it gets exercised for real.

🤖 Generated with Claude Code