test(capability-src): bring deadline-pressure and decide-escalation to 100% coverage

[enricopiovesan]

Summary

• Brings core.calculate-deadline-pressure and core.decide-escalation
  (capability-src/core-calculate-deadline-pressure,
  capability-src/core-decide-escalation) to 100% line, function, and
  region coverage (cargo llvm-cov), closing core.calculate-deadline-pressure / core.decide-escalation: bring test coverage to this repo's 100% bar #273.
• Removes genuinely dead code first (never reachable from evaluate/fail,
  confirmed by occurrence-count audit — each removed function appeared
  exactly once, its own definition, nowhere called): boilerplate copied
  from a shared template but never trimmed per-capability.
  core-calculate-deadline-pressure: skip_ws_comma,
  array_after_key_at_depth, extract_bool, parse_number_millis,
  normalize_email, trim_ascii, eq_ignore_case, ascii_lower.
  core-decide-escalation: skip_ws_comma, write_i32, extract_string,
  extract_string_at_depth, string_value_after, copy_json_escaped,
  normalize_email, trim_ascii, eq_ignore_case, ascii_lower,
  parse_ymd_days, format_score_millis (this capability never deals with
  dates or fractional scores at all).
• Adds use_case_NN_{happy,sad} tests (following the existing convention
  in both files) for every remaining reachable branch: band/decision
  boundaries, missing/malformed/absent JSON keys, unbalanced
  objects/arrays, escaped-backslash handling in the hand-rolled byte
  scanner, and non-numeric fallbacks.
• Adds a handful of direct unit tests (calling parser/formatter helpers
  like string_value_after, parse_i32, write_u32, copy_json_escaped,
  array_after_key_at_depth directly) for a few defensive branches that
  are either impossible or impractically contrived to reach through a
  realistic top-level JSON input (e.g. a buffer that's already full when a
  digit-writer is called, or a key's colon search needing to find no
  colon anywhere in the rest of the document) — standard practice for
  helper-level edge cases, and avoids fabricating unrealistic end-to-end
  inputs just to move a percentage.

Verification against the real deployed artifact

Not just cargo llvm-cov locally — ran the real
scripts/ci/gather_catalog_data.py (98 capabilities, live measurements)
and piped the output through the actual wasm32-unknown-unknown release
build of catalog-builder.wasm via wasmtime, the same pipeline
build-catalog CI runs. Both capabilities report
{'lines_percent': 100, 'functions_percent': 100, 'regions_percent': 100}
in the regenerated catalog.json.

No open issue tracks this beyond #273. capability-src/ is not governed
by any approved spec's governs list (confirmed — same as #272), so
001-registry-foundation is cited per this repo's existing precedent for
capability-src-only changes.

Closes #273.

Governing Spec

• 001-registry-foundation

Definition of Done

• Each of the two crates' evaluate() function and its helpers reach 100% line coverage.
• New #[test] cases follow the existing use_case_NN_{happy,sad} naming convention already used in both files.
• cargo llvm-cov --manifest-path capability-src/<crate>/Cargo.toml --text --show-missing-lines reports no uncovered lines for either crate.
• Live catalog badges for both capabilities show coverage at or near 100% after the next build-catalog deploy (verified via a real end-to-end pipeline run, see above).

Validation

• cargo test --manifest-path capability-src/core-calculate-deadline-pressure/Cargo.toml (36 tests, all passing)
• cargo test --manifest-path capability-src/core-decide-escalation/Cargo.toml (30 tests, all passing)
• cargo llvm-cov --manifest-path <crate>/Cargo.toml --summary-only — 100.00% lines/functions/regions for both
• cargo build --release --target wasm32-unknown-unknown for both crates
• End-to-end: real gather_catalog_data.py output piped through the real catalog-builder.wasm via wasmtime

Test plan

• CI green (build-catalog regenerates catalog.json on merge; live badges should show ~100% for both capabilities after the next deploy)

Non-goals

• The other 15 Loop-mirrored capability-src crates from Catalog: mirror Loop core.* + platform.decide-state-transition into capability-src for llvm-cov coverage #255 are out of scope, per core.calculate-deadline-pressure / core.decide-escalation: bring test coverage to this repo's 100% bar #273's own non-goals.