bcm2835-i2s: fail capture open cleanly when FIFO clear has no clock

[gifflet]

Fixes #7585

In clock-consumer (slave) mode the RX/TX FIFO clear requires PCM clock cycles to complete. When the external bit clock is absent at open time (source off/unplugged), the SYNC round trip times out, the FIFOs are left uncleared, and starting DMA on top of the stale RX FIFO makes capture free-run: DREQ sticks high, the same dead word is re-read at bus speed, and the poisoned state survives further stream cycles until reboot.

This change makes bcm2835_i2s_clear_fifos() return an error on SYNC timeout and propagates it in .prepare for the capture branch that clears stale RX data, so the open fails with -EIO instead of recording garbage. The hw_params-time clear stays tolerant on purpose: on clock-consumer links it legitimately runs before the provider's clocks are up, and the .prepare clear is the one that must succeed there.

Tested on: Raspberry Pi 4B, 6.18.34+rpt-rpi-v8, HiFiBerry Digi+ I/O (WM8804 as clock provider, dtoverlay=hifiberry-digi):

• source unplugged + capture open → clean -EIO in ~0.6 s, zero bogus data (previously: instant "success" full of DC garbage + FIFO poisoned until reboot);
• source reconnected → the very next capture is healthy (3.10 s wall for 3 s of audio), no reboot needed;
• no regressions observed: normal boots capture on the first try; playback unaffected.

[pelwell]

Replace this FIXME with a one-line explanation of the slave mode problem, e.g.

The reset will timeout in clock consumer mode without an external clock.

[pelwell]

This PR is more readable than the associated issue (#7585), and would have been sufficient on its own. Apart from my comment about the FIXME and the commit title (drop the fix(...) and add a Fixes: tag), I think this is fine.

[gifflet]

Thanks for the review! All three points addressed in the updated commit:

• FIXME replaced with your suggested one-line explanation;
• commit title reworded to kernel style (dropped the fix(...) prefix);
• added Fixes: c6aeb7de226d ("ASoC: Add support for BCM2835").

Re-tested on the same Pi 4B rig after the amend: unplugged-source capture opens still fail cleanly with -EIO and recovery on reconnect remains immediate.

[pelwell]

Thanks!