Submitting large batch of jobs fails on slow network

[randomir]

Default batch size is 20 problems. Submitting 20 full-size problems to an Advantage system/solver amounts to a single SAPI (POST) request with a 9MB payload. That's almost double the part size in the multipart upload scheme we use. Also, on a typical ADSL line, that's over a minute of upload time.

It has been observed such a slow upload request can fail with:

OSError: ('Connection aborted.', RemoteDisconnected('Remote end closed connection without response',))

[randomir]

Verified fix:

• decrease default batch size to 10 (~5 MiB for 10 full Advantage-size problems)
• ideally, limit batch size by payload bytes (default to 5 MiB)
• decouple connect from read timeout (Split request_timeout config option to read and connect timeout #440), and increase the default read timeout to 600 s
• implement retry strategy from Retry SAPI requests whenever possible #414