Delete conflict timestamp handling

[mason-sharp]

Log in spock.resolutions when we receive a DELETE that is older than
the existing row (the DELETE was processed late).

Prior to this, the DELETE would be processed and the existing row deleted.
Now we can ignore since the timestamp is older.

While we are harmonizing conflict types with Postgres, no such type
exists for a late DELETE, so a new one was created, SPOCK_CT_DELETE_LATE.
To accommodate, moved to Spock having its own conflict types, while trying
to match those in Postgres for the large overlap.

Also added spock.output_delay GUC to make internal testing easier for
this case and future testing. It optionally adds a delay to the output plugin in
the begin callback and is used by a new test that was added. That is, one
can execute SQL on one node, then execute other SQL on another while
knowing that the earlier operations will arrive afterwards and create conflicts.

[ibrarahmad]

The patch itself looks good, but we need to update the docs.

[mason-sharp]

The patch itself looks good, but we need to update the docs.

Thanks. We actually need such a section in the docs. I will do in a docs PR after doing one more conflict type handling PR soon.