fix(context): reconcile .project/ reader with CNCF spec drift (closes #372) - #374

…behavior

Feature 030 US2. Two regression tests, both outside the
`@pytest.mark.upstream` class so they run on every PR (not just
nightly):

- `test_upstream_spec_failure_message_names_both_hashes`: monkeypatches
  the tracked-hash and upstream-fetch functions to force a drift, then
  asserts the resulting `pytest.fail` message names both hashes AND
  points at `specs/030-dot-project-spec-sync/quickstart.md`. A future
  rewrite of the sync test that swallows either hash or drops the
  runbook pointer will trip this.

- `test_upstream_spec_skips_when_offline`: monkeypatches
  `urllib.request.urlopen` to raise `URLError`, then asserts
  `fetch_upstream_types_go` raises `pytest.skip.Exception` (not
  `pytest.fail.Exception`). Locks FR-007 against a future rewrite
  of the fetch path.

Also extends the sync test's failure message with a fourth block
pointing at the reconciliation runbook so the next drift-detected PR
gets a direct pointer at the runbook without hunting.

…den fixture

Feature 030 US3 + SC-002. Three new test surfaces:

- `tests/darnit/context/fixtures/full_field_coverage.yaml`: a single
  `.project/project.yaml` populated with representative values for
  every field the reader exposes today. `project_lead` uses the NEW
  list form so the scalar-or-list coercer is exercised on every CI
  run; `package_managers.docker` uses the list form for the same
  reason; `cncf_slack_channel` uses the OLD YAML key so the
  deprecation warning path is exercised (and separately asserted).

- `tests/darnit/context/test_full_field_coverage.py`: golden-dict
  comparison at the mapper boundary. `test_reader_output_matches_golden`
  loads the fixture through `DotProjectMapper.get_context()` and
  asserts the flat CEL context dict equals a hand-authored `EXPECTED`
  dict. Any silent semantic drift in a future reconciliation trips
  this. `test_extra_captures_slack_channels` verifies the NEW-IGNORED
  handling of the new `slack_channels` upstream field: raw parsed
  value lands in `ProjectConfig._extra['slack_channels']` verbatim
  and no attribute is projected on `ProjectConfig`.

- `tests/darnit/context/test_dot_project_deprecations.py`: locks both
  directions of the `cncf_slack_channel` deprecation warning.
  `test_cncf_slack_channel_emits_deprecation_warning` asserts the
  presence case emits a `DeprecationWarning` naming the old key, the
  replacement, and spec version 1.2.0.
  `test_no_warning_when_cncf_slack_channel_absent` asserts a migrated
  repo (or one that never had the field) is NOT nagged; important
  because false-positive nags erode signal quality.

Adds the full speckit trail for feature 030-dot-project-spec-sync:

- `spec.md` (with 3 clarifications recorded 2026-08-14: parse-only
  scope for new fields; one-release grace + `DeprecationWarning` for
  renamed fields; version identifier bumps 1:1 with tracked-hash file).
- `plan.md` — technical context, constitution check (PASS on all 5
  principles), project structure.
- `research.md` — Phase 0 decisions with rationale and rejected
  alternatives.
- `data-model.md` — per-field classification of every `ProjectConfig`
  attribute (KEPT / KEPT-WITH-RESHAPE / KEPT-WITH-ALIAS / NEW-IGNORED
  vocabulary that the next reconciliation reuses).
- `contracts/reader-contract.md` — public reader API contract with
  exact deprecation-warning message text.
- `quickstart.md` — maintainer runbook for THIS reconciliation and
  the next one. The upstream-sync test's failure message now points
  here.
- `upstream-diff.md` — authoritative diff summary between tracked
  hash and current upstream.
- `tasks.md` — 20 tasks, all completed.
- `checklists/requirements.md` — spec-quality validation.

Also updates the speckit plan pointer in `CLAUDE.md` and
`.specify/feature.json` to feature 030.

The test at `tests/darnit/parity/tier1/test_no_product_changes.py`
enforced feature 028's SC-006 ("parity-tests PR MUST NOT modify product
source"), but it ran on every PR that pytest collected -- so any PR
that legitimately edits `packages/*/src/` (e.g., feature 030's
`.project/` reader reconciliation) tripped the guardrail with a false
positive.

FR-014's scope is stated in its own name: "no product changes on a
PARITY-TESTS PR." A PR that does not touch `tests/darnit/parity/` is
not a parity-tests PR and the check does not apply.

Add an early skip when the diff against the base ref contains zero
files under `tests/darnit/parity/`. Feature 028's own guarantee is
unchanged: any PR that DOES touch parity tests still runs the full
grep and still fails on a product-source touch, matching what the
reviewer of darnitdevorg#370 asked for.