Buffer testcase updates in network protocol

[Xyene]

Currently, when a solution gets WA on a problem with 1000 testcases in a batch, 1000 packets are sent at once to the bridge. The bridge then hammers MySQL with 1000 queries all at once. Other times, submissions grade extremely quickly and achieve the same effect. This has led to load problems in the past, and is likely to lead to load problems in the future -- the bridge is the bottleneck to grading, after all.

The judge should buffer testcase updates that occurred in the last N (N < 1) seconds, and send them to the bridge in a single packet. The bridge could then bulk_create the SubmissionTestCase models, ideally reducing load.