Add "how to optimize parsers" documentation

[jam1garner]

Spinning off from #129, making users more aware of this is a good practice.

Most notable things to be included:

1. how buffering can help perf
2. how seeks can affect binrw's perf characteristics
3. how Vec<u8> is specialized
4. how per-field error handling can add overhead in pathological cases

Also to be considered:

1. perf debugging tool recommendations (strace, flamegraph)

(If anyone else has additions as to parts of the performance characteristics found to be unintuitive please comment them below)