Preserve voice transcription transport errors' cause and actionability

[warp-agent-staging]

Description

Voice transcription transport failures (Error: Failed to transcribe CLI voice input and its GUI sibling) were one of the noisiest client Sentry surfaces (~90k+ events, ~6k users, ~3-4k/day), because TranscribeError::Transport was a unit variant with no #[source], and TranscribeError::Deserialization was too. ServerApi::transcribe dropped the underlying reqwest::Error (any non-429 HTTP status, and any failure reading/decoding the response body) after a local log::warn!, so every event reached Sentry with an identical, causeless message and no way to tell connect vs. timeout vs. 5xx vs. a truncated body apart. Because TranscribeError was never registered via register_error!/impl ErrorExt, it defaulted to is_actionable() == true, so ordinary network flakiness paged engineering on every occurrence.

This mirrors the existing AIApiError treatment (crates/warp_errors/src/reqwest.rs, impl ErrorExt for AIApiError, and its DeserializationError split / from_transport_error hyper-source walk) that a prior PR (#14292) applied to the multi-agent path but never touched voice.

Changes

• TranscribeError::Transport now carries the source reqwest::Error instead of discarding it.
• Non-429 non-success HTTP statuses get their own ErrorStatus(http::StatusCode) variant instead of being collapsed into the unit Transport, so status is preserved.
• TranscribeError::Deserialization now carries a DeserializationError (shared with AIApiError) instead of being a source-less unit variant. TranscribeError::from_json_error classifies a reqwest::Error from Response::json(): a decode error whose source chain contains a hyper::Error (reqwest sometimes misreports a mid-stream disconnect/timeout as a decode error) or any other non-decode failure routes to Transport; a genuine decode failure routes to Deserialization(DeserializationError::Transport(_)). This mirrors AIApiError::from_transport_error.
• Added impl ErrorExt for TranscribeError + register_error!(TranscribeError):
  • Transport delegates to reqwest::Error::is_actionable() (connect/request/body/decode, 429, 5xx, staging 403 → not actionable).
  • ErrorStatus is not actionable for 5xx/429 (server-side), actionable for other 4xx (real client-side defect).
  • QuotaLimit/ServerOverloaded are not actionable (expected, user-visible conditions).
  • Deserialization(Json(_)) stays actionable (a real JSON contract bug); Deserialization(Transport(_)) delegates to the wrapped reqwest::Error.
  • Other delegates to the wrapped anyhow::Error.
• No behavior change for users: both CLI (agent_input_footer) and GUI (editor/view/voice.rs) sinks already call report_error!(anyhow::Error::new(e).context(...)) typed, unchanged; the error toast and log::warn! breadcrumbs are unchanged. Only the Sentry actionability classification (and the cause/status now present in the reported chain) changes.
• AIApiError (the multi-agent path from Preserve transport error actionability in Sentry reports #14292) is untouched.

Linked Issue

QUALITY-1231

Testing

• No tests are included: unit tests covering TranscribeError's is_actionable() classification were written and passing, then removed at the reviewer's request, along with the comments this change had added. The classification therefore has no automated regression guard.

• rustfmt --check on the touched file — clean.

• Compilation is verified by CI; the branch has been rebased onto current master.

• No UI surface change (pure Rust error-classification change), so no visual verification was needed.

• I have manually tested my changes locally with ./script/run

Agent Mode

• Warp Agent Mode - This PR was created via Warp's AI Agent Mode

CHANGELOG-NONE

[warp-agent-staging]

This PR was generated with Warp.

Comment @warp-factory on this PR to send it follow-up work.