Support TATAS-like usage in islocked and trylock spec

[tkf]

NOTICE: THIS IS NOT AN ENDORSEMENT OF SPIN LOCKS

Currently, how islocked should behave is not well-specified. This PR is an RFC for allowing usage like test-and-test-and-set lock or exponential backoff lock:

nspins = 0
while true
    while islocked(lock)
        GC.safepoint()
        nspins += 1
        nspins > LIMIT && error("timeout")
    end
    trylock(lock) && break
    backoff()
end

Notably, this can be used for judging if a given implementation of islocked is correct for APIs like semaphore (#32299) and read-write lock where multiple tasks can enter the critical section in a certain way.

From this point of view, the API could have been clearer if the name was islockable. But I think the predicate islocked can also be interpreted as "am I locked out?" or "is it completely locked?"

If we decide this API contract is a good direction, we need to fix ReentrantLock implementation (maybe in this PR or later).

[vtjnash]

Seems reasonable. I think other users have transitioned to assert_havelock or other, more correct lock usages.

[tkf]

tester_linux32 is not finished but it is hopefully OK since this PR mostly only updates the texts in the docstrings. The CI issue has been reported in #44859 and #ci-failure-help slack channel.