Following Daniels's comment (alog44001) I moved OPO refl diode back a few inches away from the focal point. That solved all of our locking problems. The error signal now crosses zero (and linear) at pump transmission peak, the I-Q plot looks symmetry and lopes now scale with power. The beam size that was hitting the diode looked like it was ~100um. Apparently that was too small. Now it's roughly a factor of 2-3 larger and all falls within the diode.

Here's a scan with ~20mW pump input to the fiber (~7mW to the OPO)

With larger error signal TTFSS UFG crossing is now a little over 100kHz. I couldn't push the gain much further without ringing something up (and boosts weren't turned on). Need to drag in a RF spectral analyzer next time.

I also put a 20dB attenuator to attenuate RF power that drives 35.5MHz EOM to check whether the 35.5MHz peak pump refl saw really came from there. SHG didn't have trouble acquiring lock so I left it there. The UGF is ~2kHz. Usually 3kHz without attenuator. 

It's a wonderful Friday....