Android CI jobs spurious failure

It seems like all of them fail occasionally but i686-linux-android is pretty bad, with an 11% failure rate per GH's metrics. Some sample messages:

adb: error: failed to copy '/checkout/target/i686-linux-android/debug/deps/cmsg-930d5ec3aabcc80f' to '/data/local/tmp/cmsg-930d5ec3aabcc80f': remote secure_mkdirs failed: Read-only file system

thread 'main' panicked at /tmp/runtest.rs:49:9:
failed to find successful test run
stack backtrace:
status: exit status: 127
stdout ---

stderr ---
/system/bin/sh: /data/local/tmp/linux_fcntl-d0799fab8ac010c3: not found

   0: std::panicking::begin_panic
             at /rustc/e16a049adbf94d610787430b6efdf31d896dc5b6/library/std/src/panicking.rs:767:5
   1: runtest::main
   2: core::ops::function::FnOnce::call_once

From https://github.com/rust-lang/libc/actions/runs/13617856722/job/38063309102 and https://github.com/rust-lang/libc/actions/runs/13659560732/job/38187528080

Start by comparing the two linked GitHub Actions runs and the failing Android CI jobs, including the i686-linux-android examples. Trace whether the adb filesystem and missing executable errors share a runner or test-execution cause. Done means the affected Android jobs no longer fail spuriously and the failure rate is measurably reduced.