Fix DeepSeek Flash model ID for OpenCode Zen

[mrubens]

Summary

• map the recommended DeepSeek V4 Flash 0731 model to OpenCode Zen's stable opencode/deepseek-v4-flash alias
• keep the dated provider slugs for OpenRouter, Vercel AI Gateway, and Baseten
• add regression coverage for each provider-specific model ID

Root cause

The dated deepseek-v4-flash-0731 slug was applied to every supported provider. OpenCode Zen serves that release through its stable deepseek-v4-flash alias, so requests using the dated OpenCode slug were rejected as unsupported.

Impact

Roomote's OpenCode Zen model recommendation now launches tasks with the provider-supported model ID. Other providers continue using their existing dated IDs.

Validation

• pnpm exec dotenvx run -f .env.test -- pnpm --filter @roomote/types exec vitest run src/model-provider-config.test.ts
• pnpm --filter @roomote/types check-types
• pnpm --filter @roomote/types lint
• pre-push oxlint, residual lint, fast type checks, and knip

[roomote-community]

No code issues found. See task

• packages/types/src/model-provider-config.ts:583 - This only changes the catalog used for future suggestion seeding. Existing OpenCode deployments retain opencode/deepseek-v4-flash-0731 in persisted model settings and selected roles; the auto-add path exits when any OpenCode model already exists (apps/web/src/trpc/commands/task-models/auto-add-models.ts:209-215), so those deployments still launch the rejected slug. Migrate persisted OpenCode DeepSeek Flash model IDs (including role/default settings) to the stable alias.
• packages/types/src/task-models.ts:228 - The alias is applied to deployment settings, but historical task-run harnessModelOverrides still bypass it. Snapshot resumes deliberately reuse the source payload, and the worker passes that raw override to OpenCode, so a pre-fix run using opencode/deepseek-v4-flash-0731 still resumes with the rejected ID. Normalize the alias when reusing persisted harness overrides (and cover snapshot resume).

_{Reviewed a1aaac3}