wasapi: address buffer overflow when using -wasapishared (#758)

## Link to GitHub Issue or related Pull Request, if one exists
Building on #745 

## Description of change
With `-wasapishared`, some users see `AUDCLNT_E_BUFFER_TOO_LARGE`
(`0x88890006`).

The existing one-device-period buffer clamp wasn't enough on endpoints
with a small shared buffer; the shared buffer is double-buffered, so a
full-buffer write often exceeds the free space.

### Fix

Added a FIFO bridge to `SharedRedirect` that decouples the game's
per-event writes from the shared engine's clock:

- **`GetBuffer`** hands the game a pointer into the FIFO tail to write
in place.
- **`ReleaseBuffer`** commits the write and drains `min(pending,
device_free)` frames to the device - so a write can never exceed what
the device accepts, structurally preventing the overflow on any
endpoint.
- **`GetCurrentPadding`** reports the FIFO fill level (capped to the
reported buffer size) so poll/timer-driven games pace correctly against
the virtual buffer.

## Testing

src/spice2x/hooks/audio/backends/wasapi/util.h

@@ -21,6 +21,10 @@ void print_format(AUDCLNT_SHAREMODE share_mode, DWORD stream_flags, REFERENCE_TI
 // IAudioClient::IsFormatSupported).
 void print_format(AUDCLNT_SHAREMODE share_mode, const WAVEFORMATEX *device_format);
 
+// scale every sample of an interleaved device buffer by `gain`, clamped to the format's range.
+// supports 16/24/32-bit PCM and 32-bit float; other formats are left untouched.
+void apply_gain(BYTE *buffer, UINT32 frames, const WAVEFORMATEXTENSIBLE &fmt, float gain);
+
 // detect IEEE float samples: WAVE_FORMAT_IEEE_FLOAT, or WAVE_FORMAT_EXTENSIBLE whose SubFormat is
 // KSDATAFORMAT_SUBTYPE_IEEE_FLOAT (Data1 == 3; _PCM has Data1 == 1)
 inline bool is_ieee_float(const WAVEFORMATEX *fmt) {