Make Follower Reads non-experimental

[rmloveland]

Currently, various parts of the Follower Reads UX are "experimental", and/or can be difficult to get working properly. There is engineering work happening to move the feature into fully-supported, non-experimental status. We will need to document the output of that engineering work.

Estimated scope of work:

• Remove experimental_ from the name of the timestamp-generating function that users are expected to call
• Document architecture updates related to this work, if any
• Document user-facing behavior changes related to this work as needed. In particular, we may need to revise the FR docs to remove the current limitations/caveats (and perhaps add others, not sure - investigation is needed).

Related issues:

• clarify that experimental_follower_reads_timestamp() is just a function that returns a timestamp #5210
• Consider removing experimental_follower_read_timestamp() from docs #6808
• Add closed timestamps to the architecture docs #6793
• Clarify that follower reads are not currently a guarantee of low latency and discuss some more factors that can affect them #5024
• Clarify replica access for AS OF SYSTEM TIME #6753

[jseldess]

@rmloveland, linking #8166 with this roadmap issue.

[rmloveland]

@andreimatei, it looks like #8816 (which you are about to review) will also close this issue, but to confirm:

AFAICT dropping the experimental_ from the function name is the only user-facing docs change that is needed to make FR "not experimental anymore". The FR-related badnesses you mention in cockroachdb/cockroach#52359 are all fairly low-level things that were/are not mentioned in the user documentation.

Do you agree with the above assessment?

[andreimatei]

Correct