fix(macos): open mic at native sample rate to prevent chipmunk audio on 96kHz devices

[ManthanNimodiya]

On macOS, microphones that default to 96 kHz (e.g. some USB/studio mics) were forced to open at 48 kHz. cpal selects the nearest supported config, which on some devices means the stream opens at 96 kHz anyway but the pipeline treats it as 48 kHz, halving the perceived duration and pitching audio up 2x (chipmunk effect).

Fix: read the device's native sample rate via default_input_config() and use that as the preferred rate instead of hardcoding 48 kHz. The downstream resampler already handles conversion to the pipeline's 48 kHz output, so no other changes are needed.

Greptile Summary

This PR fixes a pitch-shift ("chipmunk") bug on macOS microphones that natively run at 96 kHz. Instead of hard-coding 48 kHz as the preferred stream rate, the code now reads the device's actual nominal rate via default_input_config() and uses that — cpal then opens the stream at the matching rate, and the existing downstream resampler converts to the pipeline's 48 kHz output.

• Root-cause fix: preferred_rate in get_usable_device is now the device's native rate (e.g. 96 kHz) rather than a constant 48 kHz, preventing cpal from silently opening the stream at the wrong clock and treating the buffer counts as if they were 48 kHz samples.
• Safe fallback: default_input_config() failures (returns Err) fall back to unwrap_or(48_000), which restores the previous behaviour so non-96 kHz devices are unaffected.
• Settings path unaffected: when caller-supplied MicrophoneDeviceSettings include an explicit sample_rate, select_preferred_config runs first and returns early, so user overrides take full priority over the native rate.

Confidence Score: 5/5

Safe to merge. The change is a small, well-scoped substitution in a single function, with an explicit error fallback that restores the prior 48 kHz behaviour if the new device query fails.

The four-line change touches one code path in get_usable_device. The error fallback to 48 kHz means devices that don't expose a native rate are unaffected. The settings-override path (select_preferred_config) still takes full priority when a caller supplies an explicit sample rate, so no regression there. The callback estimator, resampler, and downstream pipeline all accept arbitrary input rates, making 96 kHz (or any other native rate) a valid stream configuration.

No files require special attention.

Important Files Changed

Filename	Overview
crates/recording/src/feeds/microphone.rs	Replaces the hardcoded 48 kHz preferred rate with the device's native sample rate read from default_input_config(); falls back to 48 kHz on error, preserving prior behaviour.

_{Reviews (1): Last reviewed commit: "fix(macos): open mic stream at hardware ..." | Re-trigger Greptile}

[tembo]

Nit: you can keep this strongly typed as cpal::SampleRate and avoid reaching into .0 (and potentially drop native_rate).

Suggested change

	let preferred_rate = device
	.default_input_config()
	.ok()
	.map(|c| c.sample_rate())
	.unwrap_or(cpal::SampleRate(48_000));