Frequent SPI errors

On a given drive, SPI errors will accumulate like this:

---

Normally this isn't a problem. Nearly all our communications are idempotent, and the CAN traffic that isn't is fine to drop packets occasionally, both for read and write. This allows our SPI protocol to be simple.

However, we have now run into an ECU (Chevy Bolt EPS) that is **extremely** sensitive to dropping packets, and these SPI errors result in ECU faults. https://github.com/commaai/panda/issues/2122

We have two (non-mutually exclusive) options here:
1. identify the root cause and drive down the frequency of the errors such that the probability of faulting the Bolt EPS on a given drive is very low
2. make the protocol robust to these drops, similar to USB

I'd like to avoid No. 2 if possible, so let's first try fixing No. 1 first. I suspect it's something like a race condition or we're disabling the SPI peripheral on the panda too long.