Fix relay_recent_measurement_attempt_count and friends where they are off

I've spent some time looking over the longclaw-only unmeasured relays in #40049 (closed). While there is not much concerning from the measurement side going on I think we have a bug wrt to relay_recent_measurement_attempt_count and similar values.

All of them should accumulate the attempts, results etc. of measurements done in the last 5 days (that's the current default). However, for some relays the resulting numbers are much higher than actual measurements shown in the log (see: #40049 (comment 2727670) and #40049 (comment 2727685)).

It would be useful to solve this bug as it's confusing and would help us with automating parts of the bwfile analysis when we try to spot issues early on (less weird outliers due to bugs means likely less false positives and less manual intervention).

This issue could be related to #34444 (closed). Not sure yet, though.

Oh, and shouldn't there be relay_measurement_failure_count fields where failures actually happened?