Dynamic partitioning: refactor client

[dnr]

What changed?

Refactor matching client to simplify code:

• parse task queue rpc name early and pass tqid.Partition
• move cache key construction into invokeWithPartitionCounts
• make resolvePartition return bool "loadBalance"

The only behavior difference is that matching->matching forwarded calls now
don't involve partition counts/the partition cache (there was no reason for them
to, they were directed to a specific partition).

Why?

Cleaner, more efficient code, easier to add methods.

How did you test it?

• built
• run locally and tested manually
• covered by existing tests
• added new unit test(s)
• added new functional test(s)