Perf suggestion for crlf handling

I haven't measured this but....

As long a the crlf filtering is loading the entire blob/file into a buffer, there are some opportunities here for perf gains.

In git_buf_text_is_binary() we examine the entire file. While there we could count the number of CRLFs, and plain LFs and return those sums. We should be able to use those sums in git_buf_text_gather_stats(). Or combine those 2 routines into 1. Then in git_buf_text_lf_to_crlf() add this delta "(count_crlf - count_lf)" as padding to the amount of reserved space in the destination buffer so that the memcpy loop doesn't need to realloc as the file is being converted.

Start by reading git_buf_text_is_binary(), git_buf_text_gather_stats(), and git_buf_text_lf_to_crlf() to understand their current buffer and line-ending handling. Measure the existing behavior before evaluating the suggested combined counts and padding; done means conversion avoids unnecessary reallocations without changing the resulting text.