replace pxt_powi implementation with better one

[riknoll]

Fixes microsoft/pxt-microbit#2192

We've long had an optional define (PXT_POWI) which can be used to prevent the compile from pulling in the std implementation of pow (which is 4k of code by itself). As far as I know, this flag is only actually used in the pxt-microbit to prevent hex size issues with the v1 micro:bit.

The historic PXT_POWI implementation casts the exponent to an integer which obviously provides pretty inaccurate results. This new implementation instead takes advantage of the fact that for positive non-zero x values:

Math.pow(x, y) = Math.exp(y * Math.log(x))

exp and log are already pulled into the hex, so this doesn't increase the hex file size at all (i verified this locally)

this is mostly based on @martinwork's implementation with some extra handling when x=0