ports/nrf/Makefile: Use micropython libm instead of toolchain libm

[glennrub]

Using libm from micropython reduces flash consumption about 5.5kb on nrf52
targets which have floating point support enabled.

[glennrub]

Thanks @dhalbert, for pointing out this flash optimization in PR #118.

[dhalbert]

You're welcome. And you might want to check for any stray leftover double-precision operations. Use this script and grep for __aeabi_d (__aeabi_dadd, etc.), etc. These can be large routines (like 1k-1.5k each). We had one or two that caused implied by doing some time arithmetic in a particular way (adafruit#343).

[above EDITED to make what to grep for clearer]

This is a simple caller-callee lister I swiped and adapted from Stackoverflow:

#!/bin/bash
# https://reverseengineering.stackexchange.com/questions/9113/how-can-i-generate-a-call-graph-from-an-unstripped-x86-l
inux-elf
arm-none-eabi-objdump -d $1 \
| grep '<' \
| sed -e 's/^[^<]*//' \
| sed 's/<\([^+]*\)[^>]*>/\1/' \
| awk 'BEGIN { FS = ":" } \
       NF>1 { w=$1; } \
       NF==1 && w != $1 { print "(\"" w "\", \"" $0 "\")," }' \
| sort -u