Speed-up reading of large header files

[astaric]

When a large file was passed to the -H parameter, bwa-mem2 spent more than 10 minutes just reading the lines from the file (I experienced this on a 70MB file with ~1.5M lines).

This was caused by computing the length of the already parsed header and reallocating the string for every single line in the input file resulting in quadratic complexity. The new implementation allocates a buffer that is able to fit the whole file up front and then reads all "valid" lines into the buffer before calling the original bwa_insert_header implementation. With this change the reading of header file takes <1s.