fix: enlarge target fill level of the pulseaudio playback buffer

[me-teorick]

Fixes #58

The Waydroid audio HAL requires ~10-20ms to respond to PulseAudio's PA_COMMAND_REQUEST events. Moonshine's default target fill level of the pulseaudio playback buffer (target_length) of 20ms was insufficient - the buffer would drain to zero before the audio HAL could respond to requests.

Timeline of failure:

0ms  - Audio HAL connects, writes 8,188 bytes
5-25ms - Buffer drains at 960 bytes/5ms, no requests sent yet
30ms - First PA_REQUEST sent (1,668 bytes needed)
35-40ms - Audio HAL still hasn't responded
45ms - Buffer empty, underrun occurs

After the buffer underrun occurs, sound stops working for the remainder of the streaming session. Some video playing apps freeze (waiting for sound to sync with image), other apps (notably VLC and Chrome) play silently.

The solution was to set a higher value for target_length. The new value of 60ms was determined by iterative testing of 10ms increments from the initial value of 20ms. Several tests where performed where video playing apps played correctly with sound for over one hour without any freezes or sound cuts.

[hgaiser]

Interesting issue! It's difficult for me to test, so I'll have to take your word for it :).

[hgaiser]

Would this fix still work for you if this else case is not changed? It feels strange to change the target length if there was an explicit target length requested. The default one above is semi-arbitrary, so changing it to 60msec seems fine to me.

[me-teorick]

No, I logged this some time ago, and waydroid requests target_length=4096 if I'm not mistaken, which is less than what it needs to function properly. But it does function properly when using the host's pulseaudio. So the else block is executed and the change is needed.
(I must admit that I don't fully understand that bit of code, to me it looks like the parameters that should be passed to .min() are passed to .max() and vice-versa).

[hgaiser]

Do you have any idea why it functions using the host's pulseaudio? That suggests to me we're missing some feature in our pulseaudio handling.