mormot/abouchez: new version with no shared hash table

[synopse]

The previous round was weird.
On my Intel PC, sharing the Hash table did make a performance boost. But on @gcarreno hardware, it was not faster.

We discussed about this with @OttoCoddo :
#38 (comment)
Also the alternative results #103 are not consistent with what was observer on the reference PC...
Is there some thermal throttling on the reference PC? Or some background process?

So with this commit:

• I reverted the previous code to use a per-thread hash table;
• Include a fast asm with crc32c hash computation, dedicated to our use case (branchless for up to 16 chars);
• Some other micro-optimizations.

Please run those without any command line parameters because it should guess the best settings.