fix(tauri): mobile mDNS browse falls back to instance name for node-id (#318)

[momics]

Closes #318

Problem

Mobile mDNS browse (iOS NWBrowser, Android NsdManager) hard-required a pk TXT record to accept a discovered node. Desktop/server nodes advertised via iroh's iroh-mdns-address-lookup (swarm-discovery) publish the node-id as the DNS-SD instance name (lowercase base32 endpoint id) and emit no pk TXT. Result: both mobile browsers silently skipped every desktop-advertised node — a phone could not discover a laptop/server on the same LAN.

Fix

Teach both mobile browse parsers to fall back to the DNS-SD instance name as the node-id when the pk TXT/attribute is absent, after validating it parses as a canonical iroh endpoint id. The pk-present path is untouched, so mobile→mobile and desktop→mobile discovery (which already rely on the pk TXT set by the mobile advertise path) are unchanged.

iOS — packages/iroh-http-tauri/ios/Sources/IrohHttpPlugin.swift

handleBrowseResults now extracts the DNS-SD instance name from result.endpoint first, then resolves the node-id as: pk TXT if present and non-empty → else the instance name if it validates → else skip. Added isValidEndpointId(_:).

Android — packages/iroh-http-tauri/android/src/main/java/com/iroh/http/IrohHttpPlugin.kt

onServiceResolved resolves the node-id as: pk attribute if present and non-empty → else resolved.serviceName if it validates → else return. Added isValidEndpointId(...).

Base32 validation

A node-id is a 32-byte Ed25519 public key encoded as lowercase RFC 4648 base32 without padding (confirmed in crates/iroh-http-core/src/encoding.rs — base32::Alphabet::Rfc4648Lower { padding: false }, and the same alphabet in packages/iroh-http-tauri/src/tests.rs). 32 bytes → exactly 52 chars from the a-z / 2-7 alphabet. Both helpers require length == 52 and that every char is in a-z or 2-7, rejecting malformed/non-base32 names instead of crashing.

Truncation check: the advertise side caps instance names at 63 chars (Swift args.pk.prefix(63), Kotlin args.pk.take(63)). A 52-char id is well under 63, so it is never truncated and the fallback always recovers the full id.

Verification

• CI-verified: npm run ci is green — Rust/JS build, clippy, tauri crate tests (19 + 4 + 1 doc), and tauri permissions all pass. This proves the change did not break the Rust/JS build, the tauri crate, or permissions. (The test:tauri:guest step reports vitest: command not found, a local-env gap unrelated to this change; the script still exits 0.)
• Not exercised by CI: the iOS Swift and Android Kotlin code is not compiled/run by npm run ci. There is no Swift/Kotlin unit-test harness in the plugin, so no native test was added. Correctness was reasoned by reading the surrounding native code and matching the exact on-wire encoding.

Manual cross-device repro (required before merge)

Desktop→mobile discovery of a node with no pk TXT:

1. On a desktop, start an advertiser on the LAN, e.g. via the node or deno adapter's mDNS advertise API (iroh swarm-discovery publishes the base32 node-id as the DNS-SD instance name, no pk TXT).
2. On an iOS device/simulator on the same Wi-Fi, run the tauri app and call mdns_browse_start / poll. Expect: a discovered event whose nodeId equals the desktop node's base32 id.
3. Repeat on an Android emulator/device.
4. Dial the discovered nodeId and confirm the connection succeeds.
5. Regression: confirm a mobile→mobile advertise (which sets the pk TXT) is still discovered on both platforms.

Notes

• No version-field bumps; only the two native source files changed.
• Emitting a pk TXT on desktop for symmetry is blocked upstream (0.4.0 discards custom TXT) and tracked under the Mobile iroh peer discovery parity: feed native browse results into an in-process AddressLookup #310 epic — not required here.