[roll] Roll fuchsia [audio_core] More stable micro-SRC

Treat source position errors of less than 1 fractional frame as no
error, when tuning a clock or the micro-SRC between clocks.

Also, detect when two clocks perfectly track CLOCK_MONOTONIC,
and do not actively synchronize them in that case.

Detail:
(1) Position measurements are taken in source fractional-frame
    units, then translated to nanosecond units. Position error, in
    particular, is fed into the PID feedback as a nsec-based value,
    but its precision is limited by the initial measurements and
    media-to-reference-time transformations, both of which are
    rooted in fractional-frame units. At 48kHz, a fractional frame
    is more than 2-ns, so two effectively-equivalent streams will
    still sometimes show source position error of [-2,2] ns,
    depending on the exact zx::time input. By treating deltas of
    <1 fractional-frame as 0, we won't overreact to unavoidable
    position fluctuation from source fractional-frame precision
    limitations, By keeping ns as our unit of measurement for the
    PID feedback controls, we retain additional precision for
    converging clocks and micro-SRC as rapidly as possible, once
    we DO begin rate-adjustment.

(2) This optimization applies when a device clock is in the
    monotonic domain and a custom client clock is a CLOCK_MONOTONIC
    clone that has not yet been rate-adjusted. Once we detect that
    a client monotonic-clone HAS been updated in any way (as
    revealed by generation_counter), we forever treat it as
    different (and thus rate-adjusted as any other clock).

fx test -o audio-libclock-unittests audio_core_unittests

Original-Reviewed-on: https://fuchsia-review.googlesource.com/c/fuchsia/+/589491
Original-Revision: 3f6d0e8d18fcaab85aaa195bef7d7f57e19b1ec6
GitOrigin-RevId: 34d02a524d4b5569a7cef08701e44a2fbe5e7b8b
Change-Id: I31ed3830d5c5993c3992c4f413eed12f198418db