Automatically setting shard_count on go_test based on t.Parallel() calls

[yushan26]

Bazel may not necessarily schedule tests cases under target in parallel and may assign one core to a target. This means that tests running in t.Parallel may run sequentially. To make sure that the test cases are truly parallelized on heavily loaded machines, you should instead set shard_count on the go_test target. We can automatically setting shard_count on go_test based on the number of t.Parallel() calls at the top level function. This matches what people expects under go test.