add a test about boot option descriptions

[iximeow]

ensuring we don't inflict Omicron#5112 on instances due to our own changes: boot option descriptions are load-bearing, so if we've changed them make sure we think through the consequences. i'm pleasantly surprised that EfiLoadOption just.. worked here. yay.

the procedure that gets us names we see for guest disks today is a bit subtle, but i've tried to document it fully to pair with the expectation in this test.

there are also specific conditions where failing this assertion is fine - say, if we've done #787, or changed how we're processing UEFI boot variables. just, as things are right now, we're surprisingly dependent on dancing with OVMF in exactly the same way every time a guest boots.

[gjcolombo]

This is both a good test for us to have as we (or, well, I) change how NVMe device parameters work and a great description of how our firmware image actually constructs these IDs. Assuming this still passes with the latest HEAD we should merge it!

[iximeow]

finally coming back to this: yep! gave this a run rebased on current master and the test still passes on Alpine, Debian, Ubuntu. ran it with a Windows guest on a lark and realized that this was missing a phd_skip!() for Windows guests, so i've added that. once CI's happy, in it goes.

[iximeow]

the phd-build failure in 392b4b4 is because... after i added the is_linux() clause to the new test, i rebased that small diff back onto this old branch so i'd have a nice neat tiny diff between what Greg looked at and what i'm gonna merge. is_linux(), however, is newer than this test.

so i've just rebased onto master, and it'll actually pass now 😓