highlight: apply a region's rules regardless of its start offset

[Londopy]

highlightRegion finds the region's end inside the region's own slice of the line, but compares that line-relative index against the region's absolute start offset:

endLoc := findIndex(curRegion.end, curRegion.skip, line)
if endLoc != nil {
    if start == endLoc[0] {
        searchNesting = false
    }

When the two happen to be equal, searchNesting is cleared and every rule inside the region is skipped, so the region's contents keep the region's own group.

This is visible on master today, with no other changes — in a C file:

line	result
int a;// TODO: x	TODO not highlighted
int ab;// TODO: x	TODO highlighted

The only difference is the offsets: the comment region starts at 8 and its remaining slice TODO: x is also 8 long, so the comparison matches by coincidence and the todo rule never runs. Any region with inner rules can hit this at some start offset.

Comparing against the start of the slice (endLoc[0] == 0) preserves the intent — "the region ends immediately, so there is nothing inside it to search" — without mixing coordinate systems.

Added a regression test that puts a region at every start offset on the line and checks its inner rule still applies; it fails at offset 3 before the change. pkg/highlight had no tests, so this adds the file.

Found while working on #4236, where the C #include region made it show up often (#include is exactly the length that collides with short header names).

[Londopy]

Closing this as a duplicate — I missed that #4022 already fixes it (and that #4018 tracks it). #4022 has the same endLoc[0] == 0 change plus fixes for nested-region ordering, so it supersedes this.

I've offered the regression test from here over on #4022, since that PR has no tests and the open question there was whether it breaks anything else. Sorry for the noise.