Publish says what it published

[sepehr-safari]

Closes #19. Closes #21.

publish printed nothing on stdout and exited 0 when any relay accepted any event anywhere in the run, so a script publishing several events could not tell which went out. It also waited on each relay in turn for 8 seconds of silence that any message restarted, and sent an event without checking it. This changes what it returns and how it waits.

What it returns

• stdout: each event that at least one relay accepted, as one line of JSON, once the relays have answered or its deadline has passed. What publish writes out is what was published.
• stderr: every answer with the event's id (<id>: <url>: accepted, accepted (duplicate: ...), refused: <reason>, no answer within <n> ms, closed before answering), notices with the relay's URL, and a k of n published line when something was not.
• exit: 0 only when every event was accepted by at least one relay; 1 when any was not, or there was nothing to publish; 2 when the command made no sense, including more than 32 relays.

How it waits

• Events are checked before they are sent; a bad signature is never offered.
• Relays are dialled at once, under the bound req and fetch now use.
• Each event is sent to every relay at once, and the relays share one absolute deadline to take it and answer: 10 s, --timeout to change it. A send still going at the deadline means the relay stopped reading, so it is dropped. Its reader is stopped first, since the send can be waiting for the write lock behind that reader's pong.
• Each relay has its own reader for the whole run. It answers pings, passes answers and notices over a queue, and reports a lost connection at once. An answer counts as soon as it arrives, a relay stalled mid message holds up only its own reader, and a relay that closed while publish waited on input is dialled again before the next event. An event sent down a connection that closes before answering is offered once more on a fresh one, when there is time left.
• Relay text is shown with control characters escaped, C1 included. Notices are shown up to eight per relay.

req and fetch now say when they leave relays past 32 out. The help for all three says the name lookup is the one step that cannot be cut short.

The nostr pin moves to 0.14.5 (pings and pongs arriving back to back no longer hold a read past its deadline).

Tests

Twenty-two tests drive the real command against relays on loopback, one mode per behaviour in testrelay.zig. Each of these faults, put in on its own, fails at least one of them: no signature check, printing unaccepted events, exiting 0 on any acceptance, unescaped relay text (C0, DEL and C1 separately), no resend, no redial, a stale OK counted, the notice count reset per connection. Sending to relays one after another without a bound, and joining a stuck send without stopping its reader first, each hang the suite instead of failing it; CI jobs now time out at 20 minutes so that shows up as a failure. The suite passed 15 direct runs of the test binary with no failure, the slowest 7 s.

Checked live with the release build: an event nos.lol already had came back accepted (duplicate: have this event), on stdout exactly as offered, with a silent local relay beside it dropped at 5 s.

Release

The last commit bumps the version to 0.3.0 and adds its release notes.