fix(provider): prefer explicit provider IDs for current-model filtering

[ugiya]

Summary

• prefer canonical explicit provider IDs and unique runtime aliases before model-name matching
• use model matching only to disambiguate shared runtime IDs such as google, while unknown explicit IDs fail closed
• preserve exact model-level matching for custom quota-providers
• keep intentional current-model filtering visible in verbose diagnostics without counting it as compact-status issues
• preserve the diagnostic discriminator through sanitization and persisted cache validation, with regression coverage for selection, compact output, and model-sensitive cache invalidation

This PR is intentionally limited to issue #223. It does not include the unreleased OpenCode live-model accessor or the additional formatting/reactivity changes from the local integration.

Linked Issue

Fixes #223

OpenCode Validation

• Current production released OpenCode version tested: 1.18.18
• Why this version is relevant to the fix: it supplies providerID separately from an unprefixed modelID, reproducing the identity mismatch described in the issue.

Environment

• macOS 26.5.1 on Apple Silicon (arm64)
• OpenCode 1.18.18
• opencode-quota base version 4.8.1
• Node.js 24.16.0; pnpm 11.0.0
• TUI compact status with onlyCurrentModel: true and multiple explicit providers
• Reproduction selection: { "providerID": "openai", "modelID": "gpt-5.6-sol" }

This PR is tested against the released host contract and does not depend on the proposed OpenCode live-model accessor.

Verification Performed

pnpm verify passed twice: once immediately before commit and again in the repository's pre-push hook. The gate completed Biome checks, the pinned TypeScript check, history/privacy verification, typecheck, build, the full test suite, four-surface parity, and package-content verification.

Results:

• 173 test files and 1,943 tests passed
• four-surface parity: 2 tests passed
• npm package contents: 616 files verified

Focused regression coverage verifies:

• canonical openai selects OpenAI for unprefixed gpt-5.6-sol
• unique runtime aliases such as chatgpt, codex, and chutes-ai select their catalog provider
• shared google identity is disambiguated by Gemini versus Antigravity model metadata
• unknown explicit providers and OpenRouter do not fall through to OpenAI-looking model names
• custom quota-providers retain exact provider/model matching
• intentional current-model exclusions stay in verbose diagnostics but do not become compact +N issues
• genuine provider errors still count in compact output
• the diagnostic kind survives persisted-cache validation
• changing only the selected model invalidates model-scoped custom-provider output

Manual reproduction used the OpenCode 1.18.18 TUI compact status with the environment and selection above.

Quality Checklist

• I ran pnpm run typecheck
• I ran pnpm run build
• I ran pnpm test
• This change is focused and avoids unrelated behavior changes
• I updated or added tests when behavior changed
• I updated docs when user-facing workflow, command, or config behavior changed (not required: none changed)
• For provider changes, I followed Provider Changes, or this does not apply (no provider is added)

[slkiser]

@ugiya Thanks for fix!