a67c8936

By: Tom Sydney Kerckhove <syd@cs-syd.eu>

Download the existing history in batches before syncing

Appendful's server-side read is unbounded: a fresh client sends no max-synced,
so the server materialised every command the user has into one Map and
JSON-encoded it before writing a byte. With a large history that takes longer
than warp's handler timeout, the connection is reaped, and nginx reports the
upstream close as a 502.

Rather than truncate appendful's read, which is not safe (the client derives
its cursor from max(server_id) and its own uploaded rows take the highest ids,
so a truncated read plus an upload in the same request skips rows), add a
separate endpoint for the bulk case and leave appendful for the deltas.

POST /download takes the greatest server id the client has and returns the next
batch above it in ascending id order, capped at the server's
download-batch-size (default 1024). An empty batch means the client is caught
up, so the client never needs to know the batch size. 'bevel sync' drains that
endpoint before running the appendful sync, which then only has the recent
delta to move.

Also adds a (server_user, id) index, which is what the download query wants.