Fix intermittent infinite 'Loading project...' hang (EOSE race with never-connected relays)

[dangershony]

Problem

Deep-linking to /investview/{projectId} intermittently hangs forever on the loading spinner. The indexer lookup succeeds, but the Nostr metadata fetch never completes when some relays (e.g. relay.damus.io, nostr-02.yakihonne.com) fail the WebSocket upgrade.

Root cause: the EOSE tracking snapshot in NostrCommunicationFactory.GetAllConnectedRelayNames() included all registered clients, even relays that never connected. Such relays never send EOSE, and their DisconnectionHappened event fired during startup — before any subscription was monitored — so nothing ever removed them from the tracking set and the 'all relays sent EOSE' completion callback never fired. Timing-dependent, hence intermittent.

Fixes

• EOSE tracking: only snapshot relays whose websocket is actually running (\Communicator.IsRunning); dead relays can no longer block completion (the existing disconnect-removal re-check remains as a second safety net)
• Load timeout: 30s watchdog on the InvestView remote project load — the UI now falls back to an error message instead of spinning forever
• NIP-65 via purple relays: \LookupRelayListForNPubs\ now also queries the discovery ('purple pages') relays where relay lists are actually published, with proper EOSE tracking (\includeDiscoveryRelays) and CLOSE sent on discovery relays too
• Premature close fix: \MergeProjectRelays\ now passes a real end-of-stream callback; previously with
  ull\ the relay-list subscription was closed on the first EOSE from any relay, often before results arrived
• Settings escape hatch: the standalone InvestView header (and the loading/error states) now link to /settings so users can fix relay/indexer configuration when loading fails

Validation

• Angor.Shared.Tests: 155/155 passed
• Angor.Sdk.Tests: 333 passed
• UAT (CreateProjectTest, MultiFund, MultiInvest): 3/3 passed (14m29s, real signet transactions)
• WebApp.sln builds with 0 errors