This is a response to my original post about the fuel gauge becoming unresponsive over I2C: https://e2e.ti.com/support/power-management-group/power-management/f/power-management-forum/1037566/bq27z561-not-responding-over-i2c-sometimes. My post was locked by TI, so I am posting our solution here to try and help future people who may be experiencing this issue.
After reviewing the datasheet and schematic extensively we couldn’t find any real issues, until extensive googling found this: https://e2e.ti.com/support/power-management-group/power-management/f/power-management-forum/983346/bq27z561-i2c-communication-issue
Here, the customer reports an I2C communication lockup similar to ours, depending on the state of VBAT. Although worded incorrectly, another user stated that the CE line must be 75% of VBAT to enable the device. The datasheet implies that only 1.15V is required, so driving it from a 2V IO expander should have been fine:

The simplest fix is to tie CE to VBAT. A board that was consistently locking up was modified in this way and no longer exhibited the problem.