fix(nip46): dial all configured relays instead of only the first

[oth-body]

Summary

When scanning the QR login (NIP-46 / Amber), hoot always picked relays[0] — only the first configured relay — and dialed just that one. If that single relay was slow or unreachable at scan time, the websocket connection failed and the user got an opaque error with no fallback to the other configured relays.

NIP-46 explicitly allows the nostrconnect:// URI to advertise multiple relays and lets the remote signer publish its response on any one of them. So the right fix is to advertise all of them AND listen on all of them, accepting the connection on whichever fires first.

Root cause

hoot.go's OnInitQR callback did relays := getRelayList(); relayURL := relays[0]; uri, session, err := nip46.GenerateConnectURI(relayURL, "Hoot"). The QR encoded a single relay= query parameter and nip46.Session dialed that one relay only.

When that one relay was down (relay.nostr.band times out from many networks — I verified), the user got a websocket error and login was broken even though the user's other configured relays were perfectly healthy.

Fix

nip46/nip46.go — Session now supports multiple relays throughout:

• Session.RelayURL (single string) replaced with Session.RelayURLs ([]string).
• GenerateConnectURI(relayURLs []string, appName) deduplicates the input and emits one relay= query parameter per URL via url.Values.Add. Falls back to wss://relay.damus.io if the slice is empty so pre-existing callers and tests still produce a valid URI.
• New dialRelays: parallel nostr.RelayConnect on every URL with a per-dial timeout; returns the subset that came up rather than failing the whole session on one bad relay.
• WaitForConnection subscribes for kind 24133 events on every connected relay and merges their event channels. First event wins.
• sendRequest publishes the request event to all connected relays (so it lands on whichever the signer is listening on), and subscribes for the response across all of them too.
• Close now closes every relay connection.

hoot.go — OnInitQR passes getRelayList() (the full list — falls back to the 3-relay defaultRelays when there's no relays.txt) to GenerateConnectURI instead of relays[0].

Tests

New nip46/nip46_test.go, 4 tests:

• TestGenerateConnectURIMultipleRelays — three relays in, three relay= params out, full list carried in session.RelayURLs, URI host = session.ClientPublicKey.
• TestGenerateConnectURIDedupsRelays — duplicated input relays are deduplicated in the URI output (defensive — defaultRelays already dedupes, but the helper should be robust on its own).
• TestGenerateConnectURISingleRelayFallback — empty list still produces a valid URI with the damus default.
• TestDialRelaysContinuesOnIndividualFailure — both invalid URLs return empty cleanly without panicking. This is the exact failure mode that produced the user's websocket error.

=== RUN   TestGenerateConnectURIMultipleRelays
--- PASS
=== RUN   TestGenerateConnectURIDedupsRelays
--- PASS
=== RUN   TestGenerateConnectURISingleRelayFallback
--- PASS
=== RUN   TestDialRelaysContinuesOnIndividualFailure
--- PASS
ok  hoot/nip46

Sample generated URI

go run ./cmd/showuri/ confirms the URI now carries every configured relay:

nostrconnect://ecbb4e6b927ebb4f96e57922b4a6dbd6c9166dc279b5930ab54394776ff0a6ce
    ?metadata=%7B%22name%22%3A%22Hoot%22%7D
    &relay=wss%3A%2F%2Frelay.damus.io
    &relay=wss%3A%2F%2Frelay.nostr.band
    &relay=wss%3A%2F%2Fnostr.wine

Amber (or any other NIP-46 signer) now sees all three and picks whichever it can reach; hoot listens on all three and accepts the response from any of them.