fix: bump rsfusa to v0.3.14 and make the safety-analysis pipeline real

[SoundMatt]

Summary

The rust-FuSa CLI (rsfusa) was pinned to stale v0.3.10 while the config schema, HARA schema, and content-quality rules it enforces moved on several releases. Every core safety command in CI was failing on a config parse error and being masked green by || true on every step. This PR bumps the pin to the current latest release (v0.3.14), fixes the repo's config/HARA/requirements content to satisfy the current schema, removes the masking now that the root cause is fixed, and adds real gating where the pipeline previously just logged warnings and exited 0.

• Bump rsfusa pin to v0.3.14 in ci.yml and new release.yml (single, explicit, identical pin — no floating).
• Fix .fusa.json to the current config schema (configVersion/project object/standard) — this was the actual root cause of every downstream parse failure.
• Rewrite .fusa-hara.json to the current operationalSituations/hazards/safetyGoals schema. ASIL is now derived by the tool from S/E/C against the literal ISO 26262-3 Table 4 rather than hand-typed — several hazards' previously-claimed ASIL-B didn't actually match their own S/E/C under the real table (now correctly ASIL-A/QM as applicable; the ASIL-B hazards remain ASIL-B).
• Add the 15 REQ-VIRT-013..019/REQ-MASTER-006..013 IDs that were tagged in code/tests but missing from .fusa-reqs.json and requirements.json.
• Remove || true from every safety-job step now that the config parse failure is fixed — a real regression now fails CI instead of being silently swallowed.
• Fix real check --strict findings the newer content-quality rules surface (rather than suppressing them): production-code .unwrap() → .expect("..."); test-code .unwrap() → .expect() with a rationale (rule LINT002's own remediation); reduce genuine nesting-depth hot spots (ANA002) by extracting helper functions; add #![forbid(unsafe_code)] (LINT006); annotate/fix truncating-cast sites (ANA005), including making LDF frame-ID parsing fail cleanly instead of silently truncating out-of-range IDs.
• Add the missing rsfusa coupling step — required ISO 26262-6 §6.4.4 evidence the gap report expects but CI never generated.
• Gate CI on rsfusa safety-case's missing-required-evidence warning and on rsfusa qualify's qualification badge; reorder the job so qualify/release run before safety-case/iso26262 (both check for qualify-report.json/sbom.json's presence).
• Supply rsfusa qualify real qualification metadata (--qualification-method self --qualifier ... --record-uri ...) so the badge reflects the project's actual process instead of always reading unqualified regardless of pass rate.
• Fix the coverage step's broken percentage extraction and turn the claimed "≥ 90%" comment into a real gate at the current ~85% baseline, to ratchet up over time.
• Keep rsfusa verify non-blocking — it's a real upstream rust-FuSa bug (cargo test -- --test-output=immediate, a libtest flag that doesn't exist) — but make the skip loud via a ::warning:: annotation instead of a silent || true.
• Fix a doc comment in integration_test.rs whose literal //fusa:test substring was being parsed by rsfusa trace as a malformed annotation.
• Add release.yml: attach sbom.json/provenance.json/artifact-manifest.json to the GitHub Release on publish.
• Update tool-qualification/rsfusa-tql2.md's version/qualification-method claims to match reality.

Closes #13, #14, #15, #16, #17, #18, #19, #21.

Test plan

• cargo fmt --check
• cargo clippy --all-targets -- -D warnings
• cargo test --locked (46 unit/integration + 2 doc tests)
• cargo build --release --bin rust-lin --locked
• Full local rehearsal of every safety-job rsfusa command in CI order, from a clean checkout: lint, analyze, check --strict (exit 0, 0 errors/0 warnings), check --format sarif, comp, trace, fmea, tara, cyber, coupling, boundary, hara show (0 findings), qualify (badge: self-qualified), release, safety-case (0 missing required evidence), iso26262 (13/13 objectives satisfied, 0 gaps), audit-pack — all genuinely pass, not masked.
• CI (will poll after opening)