Improve code clarity in epilogue of UTF-8 verification fast path

Improve code clarity in epilogue of UTF-8 verification fast path

The previous coding was correct, but the style and commentary were a bit
vague about which operations had to happen, in what circumstances, and
in what order. Rearrange so that the epilogue does nothing in the DFA END
state. That allows turning some conditional statements in the backtracking
logic into asserts. With that, we can be more explicit about needing
to backtrack at least one byte in non-END states to ensure checking the
current byte sequence in the slow path. No change to the regression tests,
since they should be able catch deficiencies here already.

In passing, improve the comments around DFA states where the first
continuation byte has a restricted range.