time.monotonic() was dividing a 64-bit uint by 1000.0f. This caused the compiled code toinclude _aeabi_u2lf(), which converts long long to float. In the process ul2f calls double-precision routines. I thought I had gotten rid of these by #325, but not completely.

I changed the arithmetic to break the 64-bit int into two pieces: multiply the top part by 2.0^32f and add it to the bottom part, then divide by 1000.0f. This avoids ul2f and there are no more double routines left in the image except the tiny _aeabi_d2f()`.

This gains back 3044 bytes!!