D107 - The operand solver checks that a family agrees with itself

Status: assumed

After solving, orbistoun-gen operands groups fields by family and bit position and warns when opcodes of one family read the same bits at different widths. A source field is a property of the encoding, so at most one of those readings is right - the narrow one solved without a sample high enough to need the top bit.

A warning rather than a failure: a field genuinely can differ between opcodes, and refusing to generate would make an unprovable claim in the other direction. Naming it is enough, because the cure is always the same.

Written because this failure has now occurred four times across three sittings - the wide scalar loads, then s_and_b64, s_or_b64 and s_andn2_b64 each too narrow in a different source slot. Every time it was found by a person putting the generated rows side by side, which is a habit rather than a check, and the differing-slot case is precisely what reading down a column misses.

Verified by removing the probes that fixed it and confirming the check reports both disagreements, rather than by assuming a check that reports nothing is a check that works.