feat: infer unknown lengths from context in from_buffers

[agoose77]

This PR follows up from #2718 by relaxing some rules:

• Placeholder arrays no longer need known lengths if from_buffers is able to deduce the length.
  This requirement was originally made for safety reasons — if something went wrong during rehydration etc., we would see placeholders without known lengths in contexts where they should be known. However, by imposing this restriction, the caller would need to recreate much of from_buffers to compute lengths from offset buffers. Thus, we relax this rule, and permit from_buffers to update placeholders (via a new placeholder) with new known lengths.
• Non-placeholder arrays can be encountered in contexts where the outer length is not known.
  Originally we wanted this to be an error so that we could clearly identify strange touching behaviour. However, in general it should be possible to refer to the same buffer multiple times within a form, so we should remove this rule.

[codecov]

Codecov Report

Merging #2732 (d3038d2) into main (6dc52cc) will increase coverage by 0.00%.
Report is 1 commits behind head on main.
The diff coverage is 72.41%.

Additional details and impacted files

Files	Coverage Δ
src/awkward/_nplikes/typetracer.py	76.60% <ø> (-0.03%)	⬇️
src/awkward/contents/bytemaskedarray.py	89.07% <ø> (ø)
src/awkward/contents/listarray.py	88.38% <100.00%> (ø)
src/awkward/contents/listoffsetarray.py	82.77% <100.00%> (ø)
src/awkward/operations/ak_from_buffers.py	94.07% <100.00%> (-0.08%)	⬇️
src/awkward/contents/bitmaskedarray.py	69.15% <50.00%> (-0.10%)	⬇️
src/awkward/contents/unionarray.py	85.57% <50.00%> (ø)
src/awkward/contents/regulararray.py	85.14% <50.00%> (+0.09%)	⬆️
src/awkward/contents/recordarray.py	84.55% <76.47%> (-0.28%)	⬇️

... and 1 file with indirect coverage changes

[jpivarski]

This looks very sensible. And it's a good procedure for strict rules during early development to be relaxed later as the system gets more understood.