Improving compaction heuristics + Some other tweaks while looking at compaction.

[VSadov]

The motivation for this is described in the discussion here: dotnet/runtime#115627

There is a benchmark that demonstrates that better compaction heuristics can lead to smaller heap with the throughput staying roughly the same.

This PR:

• Improves the ways in which we estimate/guess the amount of "useless" fragmentation in the heap as an input into deciding whether to do compaction.
• Implements optional eager sweep for recently allocating regions (as we expect high mortality) to improve quality of compaction, when we do compaction.
• Some improvements to reusability of free spaces for new allocations. (may help when not compacting).
• Tweaked the heuristic that decides on Gen1 budget to ignore demoted objects. Demoting from Gen2 back to Gen1/Gen0 is not an allocation.
• Some minor tweaks in affected places.

On the benchmark that started the discussion:
GCPerfSim.exe -tc 8 -tagb 200 -tlgb 2 -lohpi 0 -sohsi 50 -ramb 20 -rlmb 0.2 -sohpi 0

The commit size could be seen reduced from ~11Gb to ~7.5Gb (44% less) with throughput staying roughly the same.

For the real apps the heap size benefits will be smaller, since the benchmark intentionally exaggerates "useless" fragmentation, but we could still see some positive impact.

[VSadov]

TODO: update the net8.0 branch accordingly