refactor(sphinxdocs): use bazel label format for internal object tracking

[rickeylev]

This changes the way objects are tracked to use Bazel label format instead of dotted notation.

The dotted notation was problematic because it was ambiguous. For example, foo.bar.baz
could mean @foo//bar:baz or @foo//:bar.bzl%baz. It also required various internal hacks
to try and "unparse" the dotted notation into the "meaningful" part of an object to use
in various contexts. For example, //foo:bar usually means bar is the key term to
show, and //foo:bar.bzl%baz.qux usually means qux or baz.qux is the meaningful part.

Also:

• Make referring to things by the bzl-relative symbol names work. e.g. baz.qux can be
  used to reference //foo:bar.bzl%baz.qux
• Fix the name/referencing of runtime_env_toolchains (it was using singular in the docs)
• Add some more references to bazel inventory
• Omit --jobs=auto for sphinx_run; multi-job invocations of Sphinx make it difficult
  to interactively run Sphinx for debugging, which is the point of sphinx_run
• Add a basic test of cross-references
• Make xrefs for --name format force looking in the flag object type. This avoids finding
  objects of a matching name of another object type; e.g "precompile" is both a flag and
  attribute name.
• Cleanup the index entries: now the base name is displayed instead of the full name;
  e.g. "foo (target in //bar)" instead of "//foo:bar (target in //bar:BUILD.bazel)"