Hello all,

This patch optimizes 'get-bytevector-some'.

Currently, 'get-bytevector-some' reads a single byte at a time, doing
both 'scm_get_byte_or_eof' and 'scm_char_ready_p' on each byte,
dynamically resizing the bytevector as it goes and then shrinking it at
the end.  Needless to say, it was horrendously inefficient.

This patch changes it to simply fill the read buffer (if empty) and then
copy the contents of the read/putback buffers into a bytevector that is
never resized.

The tradeoff is that whereas the old version did everything it could to
read as many bytes as possible (even to the point of calling 'poll' at
the OS level), this new version makes no guarantees beyond what is
required by the R6RS: at least one byte will be read (or EOF).

I'd like to push this to stable-2.0.  Any objections?

      Mark