better support for network issues

[technoweenie]

• Support configurable http/tcp timeouts.

While the default timeouts are generous if you're on the same continent as the lfs servers, users should be able to customize their timeouts if they have frequent problems. Another thing to consider is whether the defaults should be increased too.

• Inspect the expires_at field for possible storage api issues.

The batch api lets servers return an optional expires_at field as a hint to the client that a 403 response may mean the credentials have expired. At the least, the client should say something (either through GIT_TRACE or STDERR). Ideally, it could call the batch api for affected files again and retry the upload automatically.

• Support configurable batch sizes

In the absence of batch api retries for storage requests with expired tokens, users should be able to turn down the batch size a bit. This will force the client to hit the LFS batch API more often to receive fresher tokens.