fix(sphinxdocs): clean stale worker output dir and doctrees on first request

[MartinNeudecker]

The Sphinx worker keeps doctrees outside declared outputs so that incremental
builds survive across invocations. A newly started worker has no digest history,
so it treats every input as changed and re-reads all docs. Re-reading against
the stale environment left by a previous worker can produce spurious warnings,
e.g. duplicate labels, which become build failures when --fail-on-warning
is enabled.

On the first request, delete the worker output directory and doctree directory
so the worker starts from a clean Sphinx environment.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[rickeylev]

Ah, I think I understand now. I was confused because bazel clears output directories, but the doctree dir isn't the output directory. It's --doctree-dir={output_dir}_doctrees (sibling of the directory). Hence its state is undefined in between worker starts or invocations.

(ideally there'd be a way to restore the state, but i think clearing things after a restart makes sense. It already does that in another case if exit code 2 comes back from sphinx)

[rickeylev]

Thanks for the fix!