cheri-cap-lib fast representability check differs from that in Sail

[PeterRugg]

As reported by Dapeng Gao:

The ‘CSetOffset’ instruction in Flute performs some optimisations (https://github.com/CTSRD-CHERI/cheri-cap-lib/blob/5c4171d3d0164d1d59ae3d6fa06ccaa4154522b8/CHERICC_Fat.bsv#L608) when computing permitted offsets, and these optimisations are not present in the Sail spec.

[PeterRugg]

Again, as reported by Dapeng:

Although the ‘CIncOffset’ and ‘CIncOffsetImm’ instructions do exactly this check, the ‘CSetOffset’ function does something slightly different. The relevant code is around here (https://github.com/CTSRD-CHERI/cheri-cap-lib/blob/5c4171d3d0164d1d59ae3d6fa06ccaa4154522b8/CHERICC_Fat.bsv#L634), where the bound comparisons are different depending on whether you are incrementing or setting the offset. This causes ‘CSetOffset’ to deviate from the spec, albeit in a ‘good’ way that allows more legal capabilities to remain tagged.