feat: schema mapping, delete fix, and gap closures CEP-6

[harsh04044]

Final PR closing CEP-6 implementation #76.

What's in this PR:

Schema-to-kind mapping table - replaced the field-presence if-else chain in handle_announcement_response() with a typed mapping table. Each entry validates field types (arrays must be arrays, strings must be strings), matching TS SDK's safeParse semantics. Rejects malformed payloads like {"tools": null}.

Delete fix - delete_announcements() now fetches existing events per kind via fetch_events() and publishes kind 5 with ["e", event_id] tags for each found event. Added fetch_events() to RelayPoolTrait and MockRelayPool.

Integration tests - roundtrip publish→discover for all 5 kinds, delete tag format verification, schema rejection tests.

Gap closures - 5 tests added for CEP-6 parity with TS SDK:

• profile metadata custom field preservation through kind 0 content
• profile metadata error resilience (no panic on publish failure)
• private server publishes relay list but not announcements
• relay_list_urls and bootstrap_relay_urls are independent
• profile metadata published regardless of is_announced_server

Also fixed a wrong integration test that expected the old ["k", kind] deletion behavior.

[ContextVM-org]

I reviewed this against the TS reference and overall this looks good to me.

The schema-to-kind mapping and the delete fix both look aligned with the intended CEP‑6 behavior, and the added parity coverage is solid.

I only have a few small follow‑ups:

• I’d add a real failure‑path test for publish_profile_metadata() (src/transport/server/announcement_manager.rs:519) by making the mock relay pool return an error from publish, just to prove we really swallow/log publication failures without panicking.
• I’d rename delete_announcements_k_tags_match_kinds() (tests/transport_integration.rs:1521) to reflect the new e‑tag behavior, since the current name still reads like the old expectation.
• As a small cleanup, I think delete_announcements() (src/transport/server/announcement_manager.rs:362) is already correct as written, but it could use the SDK’s dedicated NIP‑09 helper EventBuilder::delete() with EventDeletionRequest instead of manually assembling the kind 5 event. I don’t see this as a correctness issue, just a slightly more idiomatic rust‑nostr usage.

Nothing here feels blocking to me; these are just small cleanup/follow‑up improvements.

[harsh04044]

addressed all three - renamed the test to delete_announcements_uses_e_tags_for_published_events, replaced the no-op failure test with a real FailingPool that injects publish errors, and refactored delete_announcements() to use EventBuilder::delete() with EventDeletionRequest.