Fix SPMI to handle replays of BBINSTR jit method contexts (#41386)
Move the handling of allocMethodBlockCounts to the method context.
While this data is both an input and an output, we're more interested
in the input side.
We could also treat it as an output, if we say wanted to verify that
the replay jit wrote the same sequence of IL offsets into the BlockCount
records, but that doesn't seem crucial as changes here would also lead
to code diffs.
Fix the code that checks the AddressMap for reloc hints so it doesn't
fail if the exact lookup fails, but instead falls back to a search.
This is needed because the base of the replay count buffer is recorded
as the map key, but the instrumentation probes in jitted code refer
to offsets from this value.
Fixes #37270.