parallel replication: clean up variables determining parallelism & convert to config knobs

[problame]

Current Situation

• ZREPL_REPLICATION_EXPERIMENTAL_REPLICATION_CONCURRENCY controls overall per-filesystem concurrency
• per-filesystem PlanFS uses
  • ZREPL_REPLICATION_MAX_CONCURRENT_SIZE_ESTIMATE for dry sends (and only those)
    • note that these have both CPU requirements, thus it makes sense to limit concurrency here
  • all other code is sequential
• per-step doReplication is sequential (only 1 zfs operation at a time), and concurrency of multiple steps is controlled again by ZREPL_REPLICATION_EXPERIMENTAL_REPLICATION_CONCURRENCY

Desirable

replication:
  concurrency:
    planning: 
    replication:

Open Questions

• vars like ZREPL_ENDPOINT_MAX_CONCURRENT_SEND are used to protect a passive side from an active side that sends lots of IO / CPU-intensive requests in parallel. What to do about them?
• what about the Concurrency field in the endpoint.ListAbstractions query?

[problame]

fixed by 0ceea1b