Fall back to p2p gossip on signet when no RGS URL is configured

[hash-money]

Fixes #86.

On signet with no rgs_url configured, the gossip-source match falls through to an empty arm, leaving the wallet with no network graph at all — it can only pay invoices whose route hints bridge from its own LSP channels, and anything from a node with an announced channel fails instantly with RouteNotFound.

This falls back to set_gossip_source_p2p() on signet, matching the existing regtest behavior. The wallet's peers are exactly its LSP (and whatever the LSP talks to), so p2p gossip teaches it precisely the local topology it needs; signet graphs are small enough that sync cost is negligible. An explicitly configured rgs_url still takes precedence.

Why not RGS: no public RGS server exists for arbitrary signets, and even on Mutinynet a small LSP+mint deployment may never reach the public graph — we verified this empirically before proposing the p2p route: opened an announced bridge channel from our mint to Mutinynet's best-connected public node, watched the first-party channel get indexed within the hour, and watched our third-party mint↔LSP announcement never propagate over 15+ hours (funding UTXO unspent, valid signatures — the relaying node simply doesn't forward leaf-peer third-party gossip). Details and timed on-device repro in #86 and the linked downstream issue.

[benthecarman]

is this actually an issue? We should default to p2p gossip if no rgs url is set

[hash-money]

You're right — closing, and thanks for the pushback. I'd missed that ldk-node's builder defaults an unset gossip source to p2p (gossip_source_config.unwrap_or(&GossipSourceConfig::P2PNetwork) in build_with_store_internal), so the empty Signet/Testnet4 arms already get p2p gossip and this PR is a behavioral no-op.

Verified empirically against our signet deployment to make sure: a fresh node built with no gossip-source call, connected to our LSP, learns the island's announced channel via p2p in ~15s (the leaf sends gossip_timestamp_filter{now − 2 weeks} on connect; the LSP dumps its graph). The RouteNotFounds that motivated #86 turned out to be deployment-side — the island's gossip had gone stale (channels opened weeks earlier, zero external peering, nothing rebroadcasting, so nothing inside LDK's ~2-week freshness window to serve) — not a missing gossip source.

One small suggestion: the explicit set_gossip_source_p2p() + "We don't want to run an RGS server in tests so just enable p2p gossip" comment on the regtest arm reads as if the other empty arms get nothing — that's exactly what misled us. A one-line comment on the empty arms noting the default is already p2p would prevent the next misread; happy to repurpose this PR into that if you'd take it, otherwise closing is fine.