Further improvement to prevent potential LA related hang

[tombrazier]

Description

#25557 fixed a problem where a division by zero in the 32 bit code for Stepper::calc_timer_interval() resulted in la_interval == 0 which caused an infinite loop in the stepper ISR.

There remains a potential edge case where Stepper::calc_timer_interval() is passed a very low step_rate which results in a return value that is larger than can fit into hal_timer_t. In the very edgiest case it could even be a number that truncates to zero when converted to type hal_timer_t, also causing a hang. But even in the more general edge case it will result in a smaller value for la_interval than is correct.

This PR introduces an analogy to the 8 bit code which tests for min_step_rate and returns a fixed large number if step_rate is too small.

In addition, whilst working on this, I realised that if la_interval ends up being a large number (because step_rate is really low) then the E steps will freeze until la_interval has expired. But step_rate will only ever be really low momentarily (at the point where the E stepper reverses during a deceleration ramp due to linear advance). Long before la_interval has expired, it will receive a new, smaller value which should be honoured to avoid missed E steps. So there is a correction for this as well.

Requirements

Enable LIN_ADVANCE.

Benefits

Removes a potential edge case hang and potential lost E steps.

Related Issues

#25553
#25557