ASpace should allow spaces with size 1

[iximeow]

There was a test that explicitly rejected this, but because ASpace is exclusive it's actually fine for start == end. Adjust the test to use such a one-wide address space instead; it worked fine the whole time.

This (should?) fix #1019.

@dancrossnyc theoretically from my analysis in #1019 I believe any access to a VirtIO ISR should have immediately panic'd propolis, but I know we've both seen NICs work before and after landing #998. Do the two observations of "typically working Propolis" and "lazy-static initializer const-eval' to panic()" make sense to you? In particular, does it seem possible that a guest wouldn't touch this register except in some odd circumstance? The alternative to me seems like a guest OS/driver that we missed in testing which does touch this register when others don't?

[iximeow]

I've rebuilt a propolis-server with this patch and no longer see a call_once closure in the program with the shape "panic or unwrap failed". there's a VM on dogfood that readily hits this, though I don't know much about the guest yet. I'm going to merge this and presumably we'll see that VM not die anymore; in parallel we'll figure out what the guest/driver is that we missed in testing.

[citrus-it]

does it seem possible that a guest wouldn't touch this register except in some odd circumstance?

It would only touch this register if it was using legacy fixed interrupts as opposed to MSI-X.

[iximeow]

if it was using legacy fixed interrupts

and it turns out that an ipxe iso as one might grab from https://ipxe.org/download does this! with the patch it boots (and in my case, fails to load anything from the network), but with it reverted propolis-standalone dies quickly with

thread 'vcpu-0' panicked at lib/propolis/src/util/aspace.rs:54:9:
assertion failed: start < end
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
Abort (core dumped)