viona: FFI out of bounds fix

[dancrossnyc]

vioc_intr_poll_mq needs to have space for twice
the number of queue pairs, not the number of pairs, since interrupts are per-queue, not per-pair.

[iximeow]

importantly we'd found this by seeing propolis-standalone explode at lock decq (%rax) when ... rax == 0. the rax in question was an Arc<MemCtx>. it turns out header-check, unlike i'd thought, does not run in CI! this makes sense kind of, as we don't have a GATE_SRC to go run the tests against. but it also means that these structs were wrong and we didn't notice, and i didn't look closely because i'd assumed incorrectly that they were tested.

we've talked a bit about having header-check in CI, and i'll probably go do that next week. cloning oxidecomputer/illumos-gate and checking out stlouis in.. literally anywhere.. seems like it should be OK to do, if only on demand, or on certain files changing. maybe overriding that into a different branch when we know (like with multiqueue) we're changing the kernel too.

anyway, header-check used to fail, now passes, and i'd come up with basically the same change myself. thank you for mentioning the weird crash, 'cause it turned out to be important...