fix(models): keep gpt-5.3-codex-spark on its own display line (fix #461)

[fightingst]

Summary

• gpt-5.3-codex-spark is a distinct Codex model variant, but both
  getShortModelName() and Codex's own modelDisplayName() collapsed
  it into the base GPT-5.3 Codex label via the longest-prefix
  fallback (longest-first sort + key + '-' boundary match).
• Add explicit gpt-5.3-codex-spark entries to both SHORT_NAMES
  and Codex's modelDisplayNames, sorted before gpt-5.3-codex so
  the more specific id wins.

Repro (before fix)

CODEX_HOME=/path/to/fake codeburn today --provider codex --format json
# models: [{ "name": "GPT-5.3 Codex", "calls": 4, ... }]
# spark calls silently merged into the base line

After fix, the same fixture produces two independent line items:

{ "name": "GPT-5.3 Codex Spark", "calls": 2, "cost": ... }
{ "name": "GPT-5.3 Codex",       "calls": 2, "cost": ... }

Changes

File	+Lines	Purpose
src/models.ts	+1	Add gpt-5.3-codex-spark to SHORT_NAMES (sorted longest-first, automatically wins)
src/providers/codex.ts	+1	Add gpt-5.3-codex-spark to per-provider modelDisplayNames
tests/models.test.ts	+6	Regression: getShortModelName('gpt-5.3-codex-spark') → 'GPT-5.3 Codex Spark'
tests/providers/codex.test.ts	+12	Regression: provider.modelDisplayName('gpt-5.3-codex-spark') → 'GPT-5.3 Codex Spark'

Testing

• npm test -- tests/models.test.ts tests/providers/codex.test.ts — 126/126 pass
• E2E: wrote a fake ~/.codex/sessions/2026/06/12/rollout-*.jsonl with
  gpt-5.3-codex-spark model and ran codeburn today --provider codex --format json; confirmed spark and base now appear as separate line
  items with correct per-model cost attribution
• Verified the 4 pre-existing failures elsewhere in the suite
  (usage-aggregator, cli-proxy-path, cli-json-daily) also fail on
  main without this change — not introduced by this PR

Fixes #461