Exception raised for channels > 128

[fakufaku]

It seems the underlying libsamplerate is limited to 128 channels in the input.

# This succeeds.
data_succeeds = np.random.randn(44100, 128)
samplerate.resample(data_succeeds, 16_000 / 44_100)

# This fails with exception:
# samplerate.exceptions.ResamplingError: Channel count must be >= 1.
data_fails = np.random.randn(44100, 129)
samplerate.resample(data_fails, 16_000 / 44_100)

[tuxu]

Seems like that's a hardcoded limit in libsamplerate for sinc-methods, but not for zero-order hold / linear. We could either set CONFIG_CHAN_NR for the bundled libsamplerate or resample in chunks of 128 channels.

[fakufaku]

Right, nice find. 🙃

/* Sure hope noone does more than 128 channels at once. */
double left_calc [MAX_CHANNELS], right_calc [MAX_CHANNELS] ;

No matter what limit we set, there might be someone that hits it, so I'd go for resample in chunks of 128 channels.
We could still provide the CONFIG_CHAN_NR as a config variable so that we can use it in the bindings too.

One problem is that we will have more data to move around since the blocks of channels will not be contiguous. Do you think that's a problem?