fix(opencode): respect model limit.output config instead of capping at 32k

[JunJieLiu51520]

Issue for this PR

Closes#29363

Type of change

• Bug fix

What does this PR do?

maxOutputTokens() used Math.min(model.limit.output, outputTokenMax) which silently capped any configured output limit at the hardcoded 32k default. When users set limit.output: 384000 in their model config, the actual value sent to the API was still 32k.

Changed to honor model.limit.output when it is explicitly configured (> 0), only falling back to the 32k default when unset.

How did you verify your code works?

Traced the call path: when a model has limit.output: 384000, the function now returns 384000 instead of Math.min(384000, 32000) = 32000. When limit.output is 0 or unset, it falls back to OUTPUT_TOKEN_MAX (32000) as before.

Checklist

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

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

Potential Duplicate Found

PR #29513: fix(provider): respect limit.output config instead of silently capping at 32k
#29513

Why it's related: This PR appears to be addressing the same issue - respecting the model's limit.output configuration instead of silently capping at 32k. The title and scope are nearly identical to the current PR (#29679), just with different scope targets (provider vs opencode). Both likely fix the same underlying bug where the output token limit was being capped at a hardcoded 32k default instead of honoring user configuration.

[JunJieLiu51520]

Hi @rekram1-node 👋 — I've opened a PR that addresses this issue. Would you be able to take a look when you have a moment?

PR: #29679

[JunJieLiu51520]

Hi, I noticed that PR #29513 by @lexlian already addresses the same issue with a more complete fix (includes actual test coverage: 5 new tests + 229 passing). I wasn't aware of it when I submitted this PR — closing this one to avoid duplication.