gh-138946: Adaptive binary sort v2

[dg-pb]

Logic is as follows:

# 1. Several non-adaptive runs to estimate
#    initial `stdev` (minimal overhead)
...
# 2. Adaptive routine
while (ok < n && stdev < dynamic_threshold) {
    ...
}
# 3. Finish with the standard code (identical to what it was)

Compared to previous attempt, this has slightly less benefit for favourable cases, but minimises impact for cases where this has no benefit.

High level benchmarks.
Unwrapped float/int:

float/int wrapped in one-element list:

• Issue: list.sort enhancement proposal: Adaptivity for binarysort #138946

[tim-one]

Is your name in Misc/ACKS already? I didn't see it. I don't think it's strictly required, but the whole community appreciates the only comprehensive record we have of everyone who contributed code. So strongly encouraged.

[tim-one]

Please add more words to listsort.txt, and/or more code comments. It's not obvious what the substantial amount of new code is doing, or why. What are the implicit preconditions? Postconditions? Invariants? High-level overview of intent? Discussion of tradeoffs tried? What would you want to see spelled out if this code were brand new to you? (The single most useful question I ask myself, which made listssort.txt as wordy as it is 😉 )