[roll] Roll fuchsia [audio_core] Enable high downsampling ratios

Downsampling ratios beyond 24:1 were not previously possible, because
of the size of an internal ChannelStrip buffer in SincSampler. This
buffer must accomodate at least an entire filter-window of data, and
filter widths scale linearly with the downsample ratio.

With this fix, the filter width is also capped, leading to a graceful
decrease in quality (from gradual narrowing of effective filter width)
for downsample ratios beyond a certain threshold. Also, SincSampler
sets a slightly larger ChannelStrip buffer size, matching the new
filter-width cap specified in filter.h.

The new filter width cap is inextricably linked to ChannelStrip size,
which in turn is set so as to balance speed/quality with cost. At the
192:8 ratio, processing five frames for each cache-shift (a reasonable
performance threshold) requires a ChannelStrip buffer of 768 frames.
This width is also the exact width that guarantees at least two
side-taps during a 192:1 conversion (192K and 1K are our system max
and min frequencies, respectively). The 768-frame filter width
(expressed by a max full-quality downsample ratio of 27.5) achieves
reasonable performance at 192:8 while guaranteeing a minimum baseline
quality level at our 192:1 conversion-rate limit.

Test: fx test -o audio_fidelity_tests -- --recap
audio_mixer_profiler --frame-rates=192000:8000,192000:1000 --channels=1:1

Original-Reviewed-on: https://fuchsia-review.googlesource.com/c/fuchsia/+/494780
Original-Revision: f6d8abea7ecdd8cd3380c695709c7f068a159101
GitOrigin-RevId: b1a9c72c25f482bf0095d093eed566653c85c204
Change-Id: I7c19365e4bf8ec02f79ca40489fec3bc47a86c82