Save 5% runtime by not waiting for super low values

[alexbbrown]

Context: Reducing full scan time can make room for experimenters to put more logic into control loop. The current scan time is already great for musical use of this device.

---

Based upon 10-20 high definition runs scanning Z at 1Mhz while switching sensor, observed that while low Z values around 35 sometimes settle high, values which are nearly 0 initially do not.

This patch samples Z immediately after switching sensor and if it's low (<10) will return immediately without waiting for settling. This value could be reduced to (<2) and have nearly the same performance gain.

Alone this reduces full scan by 5%. Combined with the SPI 16-bit patch the full scan now costs 3500ms (17% improvement) when nothing is pressed.

Testing notes: Briefly tested this patch and behaviour seems normal. I will continue to observe for exceptions to this 'rule'.

[alexbbrown]

Note that if a significant fraction of the surface is above 10, then the result is more sensor reads; so available CPU time for non-scan tasks in cases where a substantial number of buttons are pressed or otherwise having a signal may be reduced slightly due to the increased read (around 20 cycles for each of the 200 buttons) followed by the read after delay.