I did the following changes which make the build significantly more robust:

• increase RAM and disk size (from original 1.5GB to: 4GB ram -> x86, 8GB ram -> x86_46, disk: 12GB)
• use own pre-defined mirror list for termux package manager to be able to disable brocken mirrors
• install (install sshd and a public key from runner) and use ssh connection to termux for less error prone scripting on android with more reliable progress and error reporting
• periodically print disk- and ram-usage during test execution to see if we run into ressource issues
• limit available memory for test execution to 3GB (ulimit) to avoid that the OS kills the sshd server process
• when using "adb shell input text" split the transmition into chunks to support longer commands.
• take and upload screenshots during startup and installation to see reason for failures during that process
• set number of cores to 4 for better performance
• switched from mac-os based runner to linux as this speeds up emulator start, build and test execution
• add timestamp to command output from ssh to detect and compare delays in execution
• updated reactivecircus/android-emulator-runner to latest verision: @v2.30.1
• close potential startup dialog "System UI isn't responding" with "wait"

All the testing took quite some time. I did't spend more time for splitting this change into smaller peaces. Hope this is fine.

Both android builds are currently red due to failing tests. For two of them I have fixes, but I will create new PRs for those: #5821 and #5940