fcof: c[gs]ethigh cross refinements

[nwf]

cheriot_instr_csethigh_cross1 and cheriot_instr_cgethigh_cross1 involve cross-products of capability {exponent, top, bottom, correction factor} and address population count. These are, mm, let's say interesting quantities, and so doubtless the correct things to consider, but will give rise to similarly interestingly shaped impossible subsets of the cross-product.

For example, it is not possible for...

• a capability with T and B both all-zeros to be anything other than correction factor vector 3'b000 (with a_mid >= B, at that) or 3'b100 (3'b111 is ruled out because a_mid cannot be smaller than all zeros).

  • all-ones T and B, on the other hand, permit more variance in a_mid: it can be >= B (by being equal), or can be <, but only if the exponent is non-zero.

• an address with fewer than 9 bits set (cp_cd_address bins valid[24] through valid[32]) to have its a_mid field be anything other than less than an all-ones B field, or greater-or-equal-to an all-zeros B field.

I'm sure there're many more such constraints. Help?