enable DYNAMIC_BMI2 by default on x86 (32-bit mode)

[Cyan4973]

Following #4251 and #4248,
it's now possible to get bmi2 working in x86 32-bit mode.

This development opens the opportunity to enable DYNAMIC_BMI2 automatically in 32-bit mode
(so far, it was only enabled for x64).

DYNAMIC_BMI2 will only capture a portion of the speed benefits of native bmi2 + avx2
as illustrated by below benchmark :

Decompression speed benchmark, measured on a i7-9700k, ubuntu 24.04, gcc 13.3.0:

dataset	-m32 dev	-m32 this PR (w/ DYNAMIC_BMI2)	-m32 -mavx2 -mbmi2
silesia.tar	861 MB/s	898 MB/s	948 MB/s
calgary.tar	836 MB/s	864 MB/s	915 MB/s
enwik7	752 MB/s	782 MB/s	822 MB/s

As one can see, DYNAMIC_BMI2 does not match the speed of native bmi2 + avx2 (although it still outperforms the absence of bmi2). This discrepancy may be attributed to the specific benefits of avx2, which likely alleviates register pressure on the x86 architecture.

DYNAMIC_BMI2 can still be manually overridden, allowing it to be disabled as needed.