chore(release): adopt voxora 0.6.1 + voxora-hf 0.6.2 → release v0.7.2

This is the release-tracking issue for the v0.7.2 patch bump of
`telora`. It adopts the two voxora patch releases that shipped on
2026-09-13:

- **[voxora 0.6.1](https://github.com/airvzxf/voxora/blob/main/voxora-hf/CHANGELOG.md#061--2026-09-13)** (coordinated,
13 crates): closes
[`airvzxf/voxora#113`](https://github.com/airvzxf/voxora/issues/113)
(`HfClient::execute_with_retry` bounded HTTP retry / backoff) and
[`airvzxf/voxora#111`](https://github.com/airvzxf/voxora/issues/111)
(streaming `Sha256` verification, bounded peak RSS).
- **[voxora-hf 0.6.2](https://github.com/airvzxf/voxora/blob/main/voxora-hf/CHANGELOG.md#062--2026-09-13)**
(single-crate patch, additive-exception path per voxora AGENTS.md
§ "Version coordination"): closes
[`airvzxf/voxora#185`](https://github.com/airvzxf/voxora/issues/185)
(advisory `flock(2)` lock on `/.lock` to serialise
concurrent `HuggingFaceSource::resolve`).

All changes are additive and API-compatible. `HfError::RetriesExhausted
{ url, attempts, last_error }` and `HfError::LockUnavailable { path,
attempts, message }` map upstream to existing `AsrError` variants; the
daemon surfaces them transparently via the existing `?`-propagation.

## Operator-visible changes

- Transient HF download failures in the daemon stderr /
`telora-models download` output now read
`retries exhausted after N attempt(s): …` (was `transport error` /
`HTTP `). Log-grep filters that key on the old strings need to
be updated.
- Worst-case additional latency before a transient failure surfaces:
~1.25 s on transient `reqwest::Error`s; up to ~60 s if the upstream
emits `Retry-After: 30`.
- Concurrent resolves against the same `(model_id, revision)` from
multiple processes / daemons are now serialised via a per-cache-dir
`flock`. Operators who run two `telora-daemon` instances sharing a
cache no longer trip over each other.

## Carried-over honest gap (still open)

- `voxora-hf` 0.6.2 still does **not** honour `ResolveOptions::max_bytes`
despite the upstream CHANGELOG claim. The 8 GiB cap telora-daemon /
telora-models wire remains inert on the HF resolve path.
Tracked via [`#148`](https://github.com/airvzxf/telora/issues/148).
- `voxora-hf` `.complete` marker is still not content-addressed.
Tracked via [`#96`](https://github.com/airvzxf/telora/issues/96).

## Issue tracker hygiene

This PR does **not** close any telora issue. The bump is purely
dependency-tracking + doc-comment refresh.

Start by inspecting the dependency declarations for voxora 0.6.1 and voxora-hf 0.6.2, then review the telora-daemon and telora-models download entry points for related documentation comments. Confirm the release metadata and comments describe the new retry and lock errors, and run the existing project checks to verify the dependency bump.

Written by the indexing model from the issue text.