Smarter deserialize_bytes and deserialize_byte_buf

[tikue]

It seems we can just read all at once rather than reading byte by byte.

[TyOverby]

This looks really good, thanks! While you are in the read_vec code, could you make a little change?

let mut buffer = Vec::new();
try!(self.reader.by_ref().take(len as u64).read_to_end(&mut buffer));

could be replaced with

let mut bytes = Vec::with_capacity(len);
unsafe { bytes.set_len(len); }
try!(self.reader.read_exact(&mut bytes[..]));

and you'd probably see even better performance.

(this implementation was ruthlessly stolen from #116)

[TyOverby]

shouldn't this be a call to visit_bytes?

[tikue]

Done.

[tikue]

Great suggestion, thanks -- done!

[dtolnay]

Still not as smart as it could be. I filed #118 to follow up in another PR.

[TyOverby]

@dtolnay: sounds good! I'll merge this and start in on the recommendations in #118