Permit tuning chunk-size when constructing a `GitOid` from a reader

[alilleybrinker]

Right now, the internal implementation of GitOid construction from a reader always reads in chunks sized to 4,096 bytes, which is the default page size on Linux for x86-64. If a user is on Linux for x86-64 and hasn't changed the default page size, then this is fine! But we really ought to be more flexible to permit tuning how reading is done to permit different chunking in case that becomes a performance problem for some users.

[alilleybrinker]

I think this would either mean:

• Use a vec instead of an array as the buffer, to permit dynamic sizing.
• Use a build script (perhaps configurable with an environment variable, or even attempting to do some host-info-snooping) to heuristically determine the best size for the read buffer.

[alilleybrinker]

We now just delegate to BufReader under the hood, and that's probably better than anything we'd write.