staging timeouts are bumped way more than they should

[xmo-odoo]

Probably because of the switch, the trigger for bumping timeouts is increased way beyond what it should: originally the intent was to set the start of the timeout on the latest build start in order to account for large runbot queues: a staging can be created then if the runbot has crazy amounts of load the build is only picked up half an hour later, which should not count in the mergebot's timeout.

So the intent was every time we receive a "pending" status (which generally marks the start of a runbot build) we bump the timeout to account for the possibility of ci/runbot or ci/l10n timing getting a late start.

However with the statuses refactoring the timeout bump now occurs any time a _compute_state results in a pending, this means not just new pending statuses but also all success statuses except for the very last one.

This is obviously not what was ever expected.