[kcrmin]

What and why

• gate8/gate9: remove the LocalStack service containers and the
  BACKTEST_TEST_SQS_ENDPOINT / LOCALSTACK_ENDPOINT_URL variables. Nothing reads
  them: tests/conftest.py and tests/test_object_store.py start their own
  LocalStack through Testcontainers. The comments now say that. Job names are
  unchanged, so required checks keep matching.
• gate3: add --cov-fail-under=77. CI measures 77.6% today (statements plus
  branches, from the develop and build(deps): cap sqlalchemy below 2.2 #103 coverage artifacts), rounded down.
  data-pipeline holds 88, so the floor should only go up.
• pyproject.toml: the comment said httpx2 does not exist. It does, and
  starlette.testclient prefers it. No dependency change: the fallback to httpx
  only warns, nothing imports httpx directly, and httpx2 shipped four releases
  in the last ten days. Swapping it in is a separate bump once a release has aged.
• test_monitor_names_the_enforcement_mechanism_actually_in_force is skipped on
  macOS, where setrlimit(RLIMIT_AS) raises ValueError. Linux CI still runs it.

Verified

• pytest -m "not docker" ... --cov-fail-under=77 (macOS): 1418 passed,
  2 skipped, total 77.55%
• pytest -m docker with no service on port 4566: 241 passed, 0 skipped
• gate9 command: 12 passed, 0 skipped
• the same docker suite with no AWS credential variables also passes (241)
• actionlint 1.7.12, ruff check, ruff format --check: clean
• Merges cleanly with fix(backtest): annotate scalar_one results #102 and build(deps): cap sqlalchemy below 2.2 #103

mypy (gate2) fails on develop itself under SQLAlchemy 2.1, with the three
scalar_one annotations that #102 adds. This PR does not touch them.

Checklist

• No job renamed and no gate weakened
• Coverage floor taken from CI's own measurement
• No dependency or runtime code change
• Related: fix(backtest): annotate scalar_one results #102, build(deps): cap sqlalchemy below 2.2 #103