docs(adr): compress to 2 records in the domain-modeling format

[lesnik512]

Rewrites docs/adr/ from 3 records (991 words) to 2 (297 words), one paragraph each, in the shape the domain-modeling ADR format asks for. Survivors are renumbered 0001-0002 in their original order and every citation moves with them.

Evaluation

Each record was tested against the admission test: hard to reverse, surprising without context, the result of a real trade-off.

Old	Outcome	New	Reason
0001 aio coverage flake fixed in the test	dropped	-	Not hard to reverse: it is two # pragma: no cover comments in one test, deleted the day coveragepy ships a fix, and tests/test_aio.py already carries the reason as a three-line comment next to them. No survivor relies on the rule, so nothing was folded.
0002 protobuf request message not a provider	kept	0001	A dependency and scope boundary: the explicit no keeps protobuf out of the runtime dependency set, and a future reader would otherwise add the provider. Its closing clause about integrations.bind vs classify_connection and its revisit trigger fold into the paragraph.
0003 streaming servicer methods must be generators	kept	0002	A public contract narrower than gRPC's and narrower than dishka's, with a silent and severe failure mode if the wrapper shape is guessed. The revisit trigger is the real boundary of the decision, so it stays as the closing sentence.

No merges: 0001 and 0002 are separate boundaries (dependency surface vs. wrapper lifetime) and share no rule.

Citations

One citation moved. docs/agents/domain.md named ADR-0001 (aio coverage flake fixed in the test) in its ADR-conflict example and listed 0001-aio-coverage-flake-fixed-in-the-test.md / 0002-protobuf-request-message-not-a-provider.md in its file-structure tree; both now name the surviving records. Nothing in modern_di_grpc/, tests/, README.md, CONTEXT.md, AGENTS.md, pyproject.toml or the workflows cited an ADR. The repo-wide sweep finds no reference to an old or dropped number.

Stale claims

Every concrete claim in the two survivors was checked against the code and holds: grpc_context_provider is providers.ContextProvider(ServicerContext, scope=Scope.REQUEST), dependencies is ["grpcio>=1.48,<2", "modern-di>=3,<4"] with protobuf only in the dev group, _build_child calls integrations.bind (never classify_connection, which does exist in modern_di.integrations), _wrap_stream_sync / _wrap_stream_aio are generators, and inject builds exactly three wrapper shapes.

One stale claim was found in the dropped record: it named the gate as the CLI flag --cov-fail-under=100, which appears nowhere in the repo. The gate is fail_under = 100 under [tool.coverage.report] in pyproject.toml, and just test-ci passes only --cov=. and the report flags. Since the record is dropped, the correction is recorded here rather than rewritten.

Gates

• pytest -q tests/test_adr_citations.py: 7 passed.
• pytest -q (full suite): 37 passed. Nothing in tests/ needs an external service; the gRPC servers are in-process on loopback.
• ruff format --check: 22 files already formatted. ruff check --no-fix: all checks passed. eof-fixer . --check: clean. No Python was touched by this change, so all three are informational.
• lychee --offline --no-progress over both new ADRs, docs/agents/domain.md, CONTEXT.md and AGENTS.md: 3 unique links, 2 OK, 0 errors.
• ty check was not run: it is part of just lint-ci, not a gate for this change, and no Python was touched.