tmc: Support sensorless homing via SPI

[rogerlz]

Summary

• When no diag0_pin/diag1_pin is set on an SPI TMC driver (2130, 2240, 5160/2160), sensorless homing polls the stallguard flag over SPI instead of a DIAG pin. The MCU polls the stallguard flag (sg2 bit of the SPI status byte) at 1kHz during homing and triggers trsync directly.
• Polling starts once the axis has reached homing speed (accel time + 50ms), since the flag is stale at standstill and noisy while accelerating.
• Host TMC register reads are held off while the poll runs, so they don't interleave with it on the bus. Daisy chains are supported.
• UART drivers keep raising the existing error.

Test plan

• Unit tests for the chain status offset and the lock handling (test/test_tmc_homing.py)
• tmc.test with no diag pin on the tmc5160, run locally with a freshly built atmega2560 dict. Not added to CI because the CI image dicts lack the new command.
• Firmware builds for stm32h723 and atmega2560 with no warnings
• Voron Trident (corexy), TMC5160 on software SPI, 0.8A, SGT 2, 60mm/s: three full G28 in a row; 30mm approaches repeat within 0.01mm (X) and 0.06mm (Y)
• TMC2130 / TMC2240, hardware SPI, daisy chain (untested on hardware)

[dalegaard]

I don't think there's any downside to always enabling this feature. Then if the user leaves out the diag pin, we register stall guard using SPI, and otherwise we keep the old functionality.

Over time, we should deprecate pin polling, as SPI is strictly better and makes auto configuration of stall guard sensitivity much easier to implement.

/Lasse