fix(opencode): set 1h prompt cache TTL for OpenRouter

[rndmcnlly]

Issue for this PR

Closes#16848

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

Adds prompt_cache_ttl: 3600 alongside the existing prompt_cache_key for OpenRouter, extending cache lifetime from 5 minutes to 1 hour. Without it, any idle gap over 5 min between prompts triggers a full cache miss and rewrite.

Tested with curl against OpenRouter -- parameter is accepted for both Anthropic and non-Anthropic models. Non-Anthropic models silently ignore it. Anthropic models show cache hits (cached_tokens > 0) on follow-up requests.

How did you verify your code works?

1. bun test test/provider/transform.test.ts -- 108 tests pass including the new one
2. curl against OpenRouter with prompt_cache_key + prompt_cache_ttl: 3600 on anthropic/claude-sonnet-4 -- confirmed cache write on first request, cache hit on second
3. curl against google/gemini-2.5-flash with same params -- 200 OK, param silently ignored

Screenshots / recordings

N/A -- no UI change.

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

[github-actions]

Thanks for updating your PR! It now meets our contributing guidelines. 👍

[rekram1-node]

Do they document this?

[rndmcnlly]

Good question -- it caught a real bug.

Yes, they document it: https://openrouter.ai/docs/guides/best-practices/prompt-caching

Key quote from the Anthropic section:

By default, the cache expires after 5 minutes, but you can extend this to 1 hour by specifying "ttl": "1h" in the cache_control object.

The documented mechanism is cache_control: { type: "ephemeral", ttl: "1h" } on message content -- not a top-level prompt_cache_ttl parameter, which is what I originally submitted. prompt_cache_ttl is silently ignored (no error, no cache write, nothing).

I ran a 3-way empirical test with a 6-minute delay past the 5-min default TTL:

Approach R1 cached R1 written R2 cached R2 written Verdict
----------------------------------------------------------------------------------------------------
A (prompt_cache_ttl=3600) 0 0 0 0 CACHE MISS
B (cache_control ttl=1h) 0 13229 13229 0 CACHE HIT
C (baseline, no TTL ext) 0 0 0 0 CACHE MISS

Only the documented cache_control approach survives. I've force-pushed a fix that adds ttl: "1h" to the existing cacheControl provider option on messages instead.

Verification script (~7 min, needs uv + OPENROUTER_API_KEY): https://gist.github.com/rndmcnlly/6680b63aeb26b5ea9ffbb56dab3030dc

Sorry for the original mistake -- my local test only checked that the parameter was accepted without error, not that it actually extended the TTL.

[github-actions]

Closing this pull request because it has had no updates for more than 60 days. If you plan to continue working on it, feel free to reopen or open a new PR.

[rndmcnlly]

Re-opened as #30190 (this PR was auto-closed by the stale bot).