fix(listener): improve error handling in session lifecycle (#1924)

- Remove .unwrap() on SessionEvent::emit in ext.rs - treat as non-fatal
- Don't treat Recorder's encode failures as fatal in post_stop
- Don't treat UI-emit failures in ListenerActor as fatal
- Make Listener's websocket tasks bail if sending ListenerMsg fails

Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

transcript store enhancements (#1879)

* feat(transcript): preserve and reindex partial hints after filter

Update transcript listener to correctly handle partial words and their
associated hints when older partial words are removed due to final
segments. Replace inline filtering logic with a two-step approach:
- collect existing partial words into a variable and filter them into
  remainingPartialWords
- build a mapping from old word indices to new indices for the kept
  partial words, then filter and remap partialHints accordingly
This prevents mismatches where hints still point to removed word
indices and ensures hints reference the correct remaining partial
word after filtering. Add a unit test that simulates partial then
final responses and asserts that remaining partial words and hints
are consistent and correctly reindexed.

* Make partial speaker hints channel-aware

Avoid cross-channel hint mis-assignment by scoping partial speaker hints to their channel. The change renames partialHints to partialHintsByChannel: a Record<number, RuntimeSpeakerHint[]>, updates initial state and all reads/writes to use partialHintsByChannel[channelIndex], and adjusts filtering/remapping and reset logic to operate on per-channel hint arrays. Tests and a component consumer were updated to read from partialHintsByChannel as well.

* Fix hint wordIndex alignment across channels

Flatten remainingWords once and reindex per-channel hints so RuntimeSpeakerHint.wordIndex refers to positions in the flattened array. Previously hints were flattened independently and kept per-channel indices, which misaligned hints for channels >0. This change computes offsets for each channel by accumulating prior partialWords lengths, adjusts each hint.wordIndex by its channel offset, and passes the flattened words and reindexed hints to handlePersist.

* Reindex flattened speaker hints by cumulative word offsets

Flattening partial hints across channels left each hint.wordIndex relative
to its original channel, producing incorrect indices when channels are
concatenated. Compute cumulative word offsets for each channel (summing the
lengths of prior channels' partialWords) and reindex each RuntimeSpeakerHint
by adding the channel's offset. Also import the RuntimeSpeakerHint type so
reindexed hints have the correct shape.

Standardize artifact naming with architecture info (#1811)

- Update desktop_cd workflow to use consistent naming:
  - macOS: hyprnote-macos-aarch64.dmg
  - Linux: hyprnote-linux-x86_64.AppImage
- Update S3 paths and GitHub release artifacts
- Add Linux download support to web app
- Update VersionPlatform type to include appimage-x86_64

This provides consistency across release and staging builds,
includes architecture information for clarity, and aligns with
Crabnebula's platform naming conventions.

Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Co-authored-by: yujonglee <[email protected]>

Add cli2 plugin wrapping official cli plugin (#1757)

Add tauri-plugin-cli and print CLI matches (#1752)

change-hook-args (#1733)

Validate Tiptap doc JSON before initializing editor (#1721)

add new icon icns (#1714)

Use tauri.conf.json version for release draft (#1711)

fix custom stt and bump