fix(google): merge timeout into caller-set http_options instead of replacing

[xlyoung]

Problem

In LLMStream._run, the request HTTP options are resolved as:

http_options = self._llm._opts.http_options or types.HttpOptions(
    timeout=int(self._conn_options.timeout * 1000)
)

When a caller passes http_options to google.LLM(...) (e.g. to set custom headers for Vertex AI Priority PayGo), the or short-circuits and uses the caller's options directly — but the caller typically doesn't set timeout. This means conn_options.timeout is silently dropped, which disables timeout-based fallback for llm.FallbackAdapter.

Fixes #5972

Fix

Merge the timeout field into the caller's http_options when they haven't set one explicitly:

http_options = self._llm._opts.http_options or types.HttpOptions()
if http_options.timeout is None and self._conn_options.timeout:
    http_options = http_options.model_copy(
        update={"timeout": int(self._conn_options.timeout * 1000)}
    )

This preserves both custom headers and framework-enforced timeouts.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 2 additional findings.

[longcw]

lgtm.