nomad: simple encode/decode benchmark

This reverts commit a687ee6d.

The LZW compression reduces the file size by ~60%, but increases
our encode and decode costs by 2x in a CPU bottlenecked path.

Job GC endpoint

LZW compress raft entries

Switch port collision checking to use bitmap instead of map

Set the size of the evaluation pool based on cores

Introduce a EvaluatePool to parallelize the Plan Applier

Make the client only reserve 512 ports.

Use compound index on node allocations to include terminal status

Client stores when it receives a task