batch several requests into one round trip

[Flatts3000]

The problem

Every request is one round trip. find_surface in Recompile's shoot_scenes.py walks down from
y=110 looking for a standing position, running two commands per candidate:

for y in range(high, low, -1):
    solid = bridge.command(f"execute unless block {x} {y - 1} {z} minecraft:air")
    clear = bridge.command(f"execute if block {x} {y} {z} minecraft:air")

That is up to a hundred sequential socket calls, each queued onto the server thread and waited on,
to answer one question.

The proposal

Let a request carry several verbs and return several replies:

{"verb": "batch", "requests": [{"verb": "cmd", "command": "..."}, {"verb": "cmd", "command": "..."}]}
{"ok": true, "replies": [{"ok": true, "output": "..."}, {"ok": true, "output": "..."}]}

The win is not raw speed so much as fewer server-thread hops: a batch can run its commands in one
server.execute rather than one each.

Notes

• Needs a decision on failure semantics: stop at the first failure, or run everything and report per
  item. Probably the latter, with an ok per reply, since a verifier usually wants the whole
  picture.
• Should refuse to batch stop, and probably screenshot, since those change which thread and
  which lifecycle stage everything after them runs in.
• Lower priority than the tier-one items: it makes existing things faster rather than making
  currently-impossible things possible.