fix(selectors): slice matcher fixes & tests

[rvagg]

Draft for now; this is fine but I need to make sure there's nothing else needed here for this piece of byte range queries (there's also more pieces).

• Don't error slice matcher when not string or bytes, silently non-match
• Cache offset in readerat so we don't over-all on the underlying read seeker
• Add lots of tests, including various edges
• Fix 32-bit breakage exposed by tests

[rvagg]

I've also removed the hard error in the slice matcher when it reaches a node that's not a string or bytes; this is more consistent with approaches of most of the other selector pieces, they just silently ignore non-matching nodes.

[rvagg]

Ready for review now, supports basic byte ranges in Trustless Spec now as this is, sans negative ranges (next PR will add that). Summary:

• Don't error slice matcher when not string or bytes, silently non-match
• Cache offset in readerat so we don't over-all on the underlying read seeker
• Add lots of tests, including various edges
• Fix 32-bit breakage exposed by tests

[hannahhoward]

LGTM

[hannahhoward]

I dunno if this matters, but this fixture doesn't quite conform to the io.Reader interface, unless i'm mistaken. https://pkg.go.dev/io#Reader

Read is supposed to read between 0 and len(p) bytes. But if len(byts[mbnrs.offset-acc:]) > len(p) that won't be the case.

[rvagg]

copy returns the number of bytes copied which is "the minimum of len(src) and len(dst)", so it's always going to be capped by len(p) in this code and we take the returned n as the authoritative offset increment. So I don't believe we're breaking the Reader contract here (unless I'm mistaken!).