Clarify permission bit encoding

[qwattash]

It is unclear how software-defined permission bits and hardware permission bits are encoded in acperm and gcperm. We should explicitly state that the instructions place the bits at the corresponding positions in the encoded capability format.

[nwf]

That's expected to be a function of the in-memory capability encoding, not an absolute across all encodings. For example, CHERIoT (my usual justification for so many things in this repo...) does not have anything that looks like a direct correspondence between an acperm/gcperm-view bit and the in-memory representation (except for the G bit, which is orthogonal to everything else).

[qwattash]

This makes sense to me to an extent. The bits defined in the base spec should have a fixed position in the acperm bit field, otherwise every implementation could move the bits around.
I think I'm fine with how it is now specified in acperm (currently in section 4.6.1).