fix(cache): preserve quantized audio sample clock coverage

[orut34iop]

Positive-duration audio packets can describe a continuous sample clock while their timestamps are quantized to a coarse container time base. For example, non-reservoir WMA frames of 2,048 samples at 44.1 kHz occupy 46/47 ms steps; treating each declared 46 ms packet independently fragments cache coverage and rejects seeks into bytes already retained.

Carry a feasible integer rational sample-clock phase across packets when the codec parameters admit exact sample geometry. Use the same coverage for the read-ahead frontier, resident ranges and retained-seek admission. A genuine missing packet, accumulated drift, missing/backward timestamps, conflicting duration or arithmetic overflow cannot be explained away by a fixed gap tolerance. Unknown codecs/packed layouts keep strict coverage. The decoded audio output clock and video admission are unchanged.

Admission covers AC-3/MPEG audio, declared AAC/E-AC-3 frame sizes, complete non-reservoir WMA v1/v2 blocks and exact PCM sample frames. This reuses upstream's duration-less successor coverage (#775) rather than presenting it as a new TrueHD fix. It overlaps #823's overflow protection in the replaced audio coverage helper: this model also checks subtraction overflow. If #823 is accepted first, preserve its regression when resolving that textual overlap. #613/#614's video interval behavior is unchanged.

Tests include 24 long sample-rate/time-base combinations, exact PCM/WMA admission, drift/real gaps/unknown layouts and actual disk and memory stores. The memory integration witnesses keep packets in the non-purgeable forward queue; they do not assume consumed history survives reclamation (#759/#819).

Validation on Apple Silicon (Mac17,3), macOS 26.7.1 / Swift 6.4:

• swift build --build-tests passed; test executable signed and verified with a valid Apple Development identity.
• Strict cooperative-pool focused run: 8 tests / 4 suites passed (including 4 real-store/gap combinations and the 24 long timing combinations).
• Full default-traits suite, serial: 795 XCTest (one pre-existing skip) and 4,619 Swift Testing tests / 636 suites passed.
• Documentation links and git diff --check passed.

CI on the initial implementation passed iOS/tvOS/visionOS builds and default/all-traits tests, including both strict cooperative-pool runs. Its later standalone regression step exposed a missing source entry in test-software-packet-read-ahead.sh; the follow-up commit adds SoftwareAudioPacketCoverage.swift to that build. All six standalone packet regressions now pass locally (packet/video coverage, disk FIFO, read-ahead, read admission and stored packets), with Apple Development signatures verified. CI for the follow-up commit is pending. These are deterministic timing and store witnesses, not new physical media/device acceptance.