Minimal change for unbounded ranges

[CAD97]

Closes #15

This is what I believe to be the minimal diff to support right-unbounded ranges with TextSize/TextRange. TextSize(u32::MAX) is reserved to mean ∞/unbounded, and the only functions that outright need to be adjusted to respect those semantics are checked_add (just replaced with saturating_add for now), len (return TextSize::INF for unbounded ranges), and the actual indexing.

This added semantics to TextSize does make the following impls questionable, and deserving of a new fresh look:

• From<u32> for TextSize: u32::MAX is lossy
• From<TextSize> for u32: TextSize::INF is lossy
• TryFrom<usize> for TextSize: u32::MAX is lossy
• From<TextSize> for u32: TextSize::INF is lossy

[CAD97]

As for constructors taking advantage of this, I think a not-horrible option might be fn TextRangeTo(_: TextSize) -> TextRange and fn TextRangeFrom(_: TextSize) -> TextRange. Definitely... not traditional, but work to get the point across.

[matklad]

Not a fan of it, for roughly the same reason I was not a fan of reversed ranges: this does make one corner case significantly better, but it also expands the (conceptual) set of values a TextRange can have, making the invariants less strict. And that makes the common case slightly more awkward. But, because the common case is so much more common, I feel like overall this adds to awkwardness.

I think usize::from(ts).. and ..usize::from(ts) should be fine. We can also add more targeted hacks like s[ts.suffix()] and s[ts.prefix()] (the hacks returning std::ops::Range{To,From}<usize>).