scripts: add full-drive test helper for macOS

[sbrunner-atx]

What

Adds scripts/f3-drive-test-macos.sh: a helper that runs a complete erase / f3write / f3read verification cycle on macOS, in the spirit of the existing log-f3wr script but handling the macOS-specific pitfalls.

Why

Running f3write/f3read by hand on macOS has several traps this script closes:

• diskutil disk numbers are assigned by connection order, so when testing several drives it is easy to erase the wrong one, or to attribute a log to the wrong physical drive. The script requires a user-chosen drive label, records it together with the USB location ID and device serial number, and asks for confirmation naming that label before erasing. It also refuses internal disks and partition slices outright.
• Reading back immediately after writing measures the page cache, not the drive. The script unmounts and remounts between the passes.
• Some failing drives only show their defect under repeated or thermally loaded reads. The script supports --reads N consecutive passes and a --cooldown / --replug pause between write and read, so failure time can be compared across passes.

It needs only f3write/f3read (from PATH or a sibling build/ directory), so it works with the stock macOS build and is independent of #283.

Origin

This came out of diagnosing a batch of refurbished 128 GB drives: one unit wrote all 114 GB cleanly but could not sustain more than about 90 seconds of sequential reading before disconnecting from the bus, with time-to-failure shrinking on consecutive passes (about 100 s cold down to about 9 s warm) - a clear thermal signature that a single default write/read pass does not isolate.

Disclosure: developed with AI assistance; reviewed and field-tested by the author on real hardware.

[sbrunner-atx]

Field report from the first production use of this script: I tested a 5-pack of refurbished SanDisk Ultra 128 GB drives from eBay. Two of the five failed in the same way - the full 114 GB write completes fine, then the drive disconnects 37-60 seconds into the read-back ("Device not configured").

The repeated read passes (--reads) turned out to be the key: time-to-failure shrinks from about 100 s cold to about 9 s warm across back-to-back passes - a thermal signature that a single write/read pass does not isolate. The good drives throttle their read speed to ~77 MB/s and run forever; the defective ones read at 144-160 MB/s, never throttle, and cook themself into a bus dropout.

The descriptor logging also earned its place: the two defective units are a different hardware revision, identifiable without any test - media name "Ultra" with serial prefix 0401, while all good drives report "SanDisk 3.2Gen1" with prefix 0901. The correlation held on every drive checked.