safe_secondary_reads_causal_consistency.js is not robust to NetworkInterfaceExceededTimeLimit errors

The test expects a find command to fail with either MaxTimeMsExpired or StaleConfig. It's possible to fail with NetworkInterfaceExceededTimeLimit if mongos hits the MaxTimeMs at the time that it attempts to get a connection to the shard because NetworkInterfaceTL uses the timeout set on the request as the timeout for grabbing a connection (and the timeout set on the request is set to the remaining maxTimeMs). This can of course happen outside of this test, so this ticket is to investigate/make a decision about whether we just want to change this test to accept NetworkInterfaceExceededTimeLimit errors, or if we'd want to do something like convert NetworkInterfaceExceededTimeLimit errors into MaxTimeMsExpired errors if the reason we hit NetworkInterfaceExceededTimeLimit is because we hit MaxTimeMs.

Acceptance criteria: determine if throwing NetworkInterfaceExceededTimeLimit, and replacing MaxTimeMsExpired, is appropriate. If not, avoid replacing the error.