Sweep-cleanup of CI workflows, templates, and docs for references to
microcrates and micro-* feature flags that have been deleted or are
queued for deletion via absorption PRs #52 #54 #55 #56 #57 #58.

Changes:
- .github/workflows/ci.yml - removed deleted features (micro-lock,
  micro-plan, micro-policy, micro-process, micro-store) from BDD matrix
- .github/workflows/mutation.yml - removed shipper-plan, shipper-policy,
  shipper-levels from the mutation-testing target list
- .github/workflows/release.yml - added note that publish order is
  finalized in Phase 8 of the decrating plan
- templates/circleci-config.yml - matching feature-matrix cleanup
- docs/architecture.md - per-crate absorbed notes in microcrates table
  plus top-of-section notes on Dependency Graph and Module
  Responsibilities pointing out those sections reflect pre-decrating state
- docs/testing.md - removed deleted crates from example test invocations
  and the mutation-testing example
- RELEASE_CHECKLIST_v0.3.0.md - header note listing absorbed crates and
  pruned publish steps
- RELEASE_NOTES_v0.3.0.md - reworded Modular Architecture bullet to
  reflect the consolidated public-crate layout

References to still-in-flight absorptions (auth, environment, git, storage,
engine-parallel, registry, progress) are left alone - they will be cleaned
up in their respective absorption PRs.

Per docs/decrating-plan.md.

Sweep-cleanup of CI workflows, templates, and docs for references to
microcrates and micro-* feature flags that have been deleted or are
queued for deletion via absorption PRs #52 #54 #55 #56 #57 #58.

Changes:
- .github/workflows/ci.yml - removed deleted features (micro-lock,
  micro-plan, micro-policy, micro-process, micro-store) from BDD matrix
- .github/workflows/mutation.yml - removed shipper-plan, shipper-policy,
  shipper-levels from the mutation-testing target list
- .github/workflows/release.yml - added note that publish order is
  finalized in Phase 8 of the decrating plan
- templates/circleci-config.yml - matching feature-matrix cleanup
- docs/architecture.md - per-crate absorbed notes in microcrates table
  plus top-of-section notes on Dependency Graph and Module
  Responsibilities pointing out those sections reflect pre-decrating state
- docs/testing.md - removed deleted crates from example test invocations
  and the mutation-testing example
- RELEASE_CHECKLIST_v0.3.0.md - header note listing absorbed crates and
  pruned publish steps
- RELEASE_NOTES_v0.3.0.md - reworded Modular Architecture bullet to
  reflect the consolidated public-crate layout

References to still-in-flight absorptions (auth, environment, git, storage,
engine-parallel, registry, progress) are left alone - they will be cleaned
up in their respective absorption PRs.

Per docs/decrating-plan.md.

Fixes two distinct failures that have kept CI red on `main` for weeks
and block every open PR:

1. Snapshot mismatch in `publish_invalid_manifest_content_snapshot`
   (`crates/shipper-cli/tests/e2e_expanded.rs`). On Linux CI runners,
   `cargo metadata` reports the invalid manifest using a path relative
   to the workspace CWD (e.g. `../../../tmp/.tmpXXX/Cargo.toml`). The
   tempdir replacement consumed the leading `/`, leaving `..` fragments
   that the previous single-pass `.replace("../", "")` could not clean
   up (it left `..<TMPDIR>/...`).

   `normalize_stderr` now:
   - converts backslashes to forward slashes so path separators are
     platform-independent in snapshots; and
   - iteratively strips `../` and any residual `..<TMPDIR>` fragments
     until stable.

   Updated the committed snapshot to the canonical forward-slash form.

2. Race in `FileStorage::write` causing
   `tests::concurrent_reads_and_writes_same_file` to panic with
   "No such file or directory (os error 2)" on macOS and Linux. All
   threads writing to the same destination shared a single
   `<basename>.tmp` scratch file, so one thread's `rename` could
   remove the file while another thread's `rename` was still pending.

   Each write now uses a unique temp filename built from pid, thread
   id, and subsecond timestamp. Adjusted two local tests that inspected
   the old fixed `.tmp` name to observe the more general invariant
   ("no `.tmp` extension files remain after a successful write") and
   relaxed `atomic_write_overwrites_stale_tmp_from_prior_crash` to only
   assert the new write succeeds (the new impl intentionally does not
   clobber unrelated stale `.tmp` files).

Validated locally on Windows:
- `cargo test -p shipper-storage -- --include-ignored` → 93 lib + 33
  integration + 1 doctest pass (including
  `concurrent_reads_and_writes_same_file`, previously `#[cfg_attr]`
  ignored on Windows; now runs cleanly).
- `cargo test -p shipper-cli --test e2e_expanded` → 126/126 pass.
- `cargo fmt --all -- --check` and
  `cargo clippy -p shipper-storage -p shipper-cli --all-targets
  --all-features -- -D warnings` both clean.

Coverage-job-only validation (`cargo llvm-cov`) is deferred to CI.

Unblocks green CI for the decrating PR queue (#48-#58).