feat: add persistent worker for sphinxdocs (#2938)

This implements a simple, serialized persistent worker for Sphinxdocs
with several optimizations. It is enabled by default.

* The worker computes what inputs have changed, allowing Sphinx to only
rebuild what
  is necessary.
* Doctrees are written to a separate directory so they are retained
between builds.
* The worker tells Sphinx to write output to an internal directory, then
copies it
to the expected Bazel output directory afterwards. This allows Sphinx to
only
  write output files that need to be updated.

This works by having the worker compute what files have changed and
having a Sphinx
extension use the `get-env-outdated` event to tell Sphinx which files
have changed.
The extension is based on https://pwrev.dev/294057, but re-implemented
to be
in-memory as part of the worker instead of a separate extension projects
must configure.

For rules_python's doc building, this reduces incremental building from
about 8 seconds
to about 0.8 seconds. From what I can tell, about half the time is spent
generating
doctrees, and the other half generating the output files.

Worker mode is enabled by default and can be disabled on the target or
by adjusting
the Bazel flags controlling execution strategy. Docs added to explain
how.

Because `--doctree-dir` is now always specified and outside the output
dir,
non-worker invocations can benefit, too, if run without sandboxing. Docs
added to
explain how to do this.

Along the way:

* Remove `--write-all` and `--fresh-env` from run args. This lets direct
  invocations benefit from the normal caching Sphinx does.
* Change the args formatting to `--foo=bar` so they are a single
element; just
  a bit nicer to see when debugging.


Work towards https://github.com/bazel-contrib/rules_python/issues/2878,
https://github.com/bazel-contrib/rules_python/issues/2879

---------

Co-authored-by: Kayce Basques <kayce@google.com>
Co-authored-by: Richard Levasseur <rlevasseur@google.com>