Rare off-by-one edge case in header-aware visitor

[airbreather]

Cursively/src/Cursively/CsvReaderVisitorWithUTF8HeadersBase.cs

Lines 473 to 478 in eaf568a

	private unsafe void VisitHeaderChunk(byte* b, int byteCount, bool flush)
	{
	int charCount = _headerDecoder.GetCharCount(b, byteCount, flush);
	if (_headerBufferConsumed + charCount <= _headerBuffer.Length)
	{
	fixed (char* c = &_headerBuffer[_headerBufferConsumed])

If the input chunks were cut precisely before the byte that denotes the end of a header whose length is exactly equal to the configured max header length, and everything's valid UTF-8, then this method will be called one last time with an empty input.

_headerBufferConsumed will be equal to _headerBuffer.Length, which is actually supposed to be OK when charCount is 0, since we're not actually going to write anything into the array... but since that index is outside the array, it actually throws an exception.

[airbreather]

Found this when updating the test suite to improve coverage for the header-aware visitor.