Fix inflate match output bound check

[HDauven]

Fixes #200.

Since decompress_with_limit was added, this check has compared an absolute end position with the remaining output capacity. This can send matches through WriteLenBytesToEnd even when the whole match fits.

Compare the match length directly with the remaining capacity. The separate overlap check can also go because apply_match already handles the same case through transfer.

The overall performance difference is small since this sits outside the main inflate loop, but the check is now correct and avoids unnecessary state-machine work.

[oyvindln]

This is probably the correct direction but will have to look over the full issue to see whether there is something else missing here.

Also something to look at performance-wise is how changing those checks affects the ability for the compiler do optimize out bounds checks later in that "switch case". Since we don't use unsafe, helping the compiler evade bounds checks and thus reducing branches and emitting code for panics in hot paths is important.