Fix Android USB missed-cycle stalls

[businesscurry123]

Summary

Fixes #46.

• Bound Panda::can_receive() CAN bulk reads to 5 ms instead of using the repo default TIMEOUT=0.
• Made the boardd_can_recv 100 Hz deadline arithmetic signed and added recv=<ns> timing to missed-cycle logs.
• Added docs/android_usb_host_missed_cycles.md with the Android no-root USB path, root-cause analysis, and a field checklist for ROM/cable/power/scheduler diagnosis.
• Added scripts/android_usb_host_missed_cycles_triage.sh, which can collect a diagnostic bundle either directly inside Android/Termux or from a desktop through adb.
• Linked the guide from README.md.

Why This Targets the Blocking Path

This PR focuses on the USB receive path that can actually consume the 10 ms boardd_can_recv loop budget on Android no-root setups. Timing reset alone is not enough here because the current loop already resets next_frame_time = cur_time when it falls behind; if Panda::can_receive() can still block indefinitely inside libusb_bulk_transfer(..., timeout=0), the next cycle can miss again for the same reason.

The added recv=<ns> log makes that distinction visible in field reports: if missed cycles correlate with CAN receive time, the Android USB host/ROM/cable/power path is the suspect rather than only scheduler catch-up arithmetic.

Root Cause

boardd_can_recv has a 10 ms loop budget. On Android no-root, pandad.py launches boardd through termux-usb, and boardd wraps the Android USB permission file descriptor with libusb_wrap_sys_device.

Before this PR, Panda::can_receive() called:

usb_bulk_read(0x81, data, RECV_SIZE);

That used the default TIMEOUT, which is 0 in this repo. libusb documents a zero synchronous transfer timeout as unlimited, so a slow Android USB host/ROM path can hold the receive loop longer than the 10 ms cycle budget and trigger missed cycles.

The existing usb_bulk_read() and can_receive() logic already treats timeout/no-data as healthy, so bounding the CAN receive read is compatible with the current behavior while preventing one USB read from blocking indefinitely.

Testing

• git diff --check
• git diff --cached --check
• bash -n scripts/android_usb_host_missed_cycles_triage.sh
• Ran scripts/android_usb_host_missed_cycles_triage.sh .triage-test locally and verified it produced the output directory, then removed the test output.
• Reviewed the touched C++ paths and the Android no-root termux-usb launch path.

I could not run the SCons build in this local Windows checkout because scons and a C++ compiler are not installed here.