* DOC: Fixes for 18 broken links

This, with PR #16465, should fix nearly all the remaining broken links
on the site. 4 or 5 others should be easy to fix and just
need attention from someone more knowledgeable -- will
open an issue. For release notes with dead links,
I could usually find links on archive.org for roughly contemporary
versions.

* DOC: Update to "Fixes for 18 broken links #16472"

* Obsolete links, previously commented out, now deleted:
https://github.com/numpy/numpy/pull/16472#discussion_r433928958

* Semantic markup for reference to Python class:
https://github.com/numpy/numpy/pull/16472#discussion_r433553928

* Missing :ref: in internal link:
https://github.com/numpy/numpy/pull/16472#discussion_r433554484

Not included: Resolution on using external/internal doc link in .py:
https://github.com/numpy/numpy/pull/16472#discussion_r433554824

* DOC: Add internal link for 'Fixes for 18 broken links' PR #16472

Making reference [1] an internal link in function_base.py => numpy.vectorize.html

* DOC: Redirect 2 link fixes in PR #16472

* governance.rst link reverted
* ununcs.rst `overridden` link goes where it was meant to


per https://github.com/numpy/numpy/pull/16472#pullrequestreview-424666070

DOC: Add axis to random module "new or different" docs

Limiting the number of loops performed in the %timeit evaluation
results in a ~20% speedup in the time it takes to build the
documentation.

The results of the "benchmarking" are less accurate from the limited
number of evaluations, but the relationship between the timing
between the legacy and Generator sampling methods is preserved.

Added `permutation` to `choice` and `shuffle` in the list of
highlighted methods with an `axis` kwarg.

Modified the introductory/overview documentation for the random
module to highlight the addition of the `axis` kwarg to
methods like `shuffle` and `choice` for dealing with
multi-dimensional arrays.

Related to gh-16075