fix(quota): normalize ambiguous usage count semantics

[SaladDay]

Summary

• normalize the ambiguous *_usage_count fields against the server-provided *_remaining_percent
• support both legacy responses where the reported count means remaining and newer responses where it means consumed
• fall back to the authoritative percentage when neither count interpretation agrees
• make quiet output use the same normalization as the text dashboard
• keep --output json and SDK responses raw and unchanged

Why

The quota API has returned both of these shapes over time:

• total=3, usage_count=3, remaining_percent=100 (reported count behaves as remaining)
• total=5, usage_count=0, remaining_percent=100 (reported count behaves as consumed)

The text renderer treated the field as remaining while quiet mode treated it as consumed, so the same payload could produce contradictory output. This change uses the explicit remaining percentage to disambiguate the count and avoids presenting a count when the fields cannot be reconciled.

Validation

• npx bun test test/utils/quota.test.ts test/output/quota-table.test.ts test/commands/quota/show.test.ts (19 passed)
• MMX_CONFIG_DIR=<isolated> npx bun test (411 passed)
• tsc --noEmit
• ESLint on changed files
• npx bun run build:dev

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}