L3 fixes

[osresearch]

This patch fixes #235 and #236 to correctly validate L3 commands with non-zero jerk2 and also flips the gJerk1 and gJerk2 globals if the sign on gSteps1 or gSteps2 is negative in legacy mode.

[EmbeddedMan]

The change from gJerk1 to gJerk2 is clearly a type (sorry!) and I'm sure that's correct.

The other two changes feel right, but I don't have a deep enough understanding of the mathematics to know that they are correct. @oskay do you think negating the Jerks is correct?

[osresearch]

I'm also not positive that the inversion of vel/accel/jerk if steps is negative is right in any case, but it is the documented behaviour for LM so it is odd that L3 doesn't do it for jerk as well.

Normally the sign of the Rate parameters are used to indicate initial motor direction. When Rate signs are used for direction, then Steps parameters should be positive. However, if the sign of Rate is positive, then the sign of the Steps parameters can be used to control initial motor direction instead. Internally, if an axis has a negative Steps and a positive Rate, the EBB code will flip the sign on Rate and Steps as well as Accel to make the math work out.

[EmbeddedMan]

I agree - it would feel really weird that this same step should only be in one of those two commands. If it's in LM it feels like it should be in L3. But @oskay is the real mind behind all of the math here, so I trust his judgment on if this change should be made or not.