HSv3 descriptors rejected because of bad SRV start time computation

When we introduced legacy/trac#25552 (moved), we started OPE encrypting the time diff since the start of the SRV run. We also have some logic on which SRV period we should use to calculate the time diff:

  if (is_current) {
    srv_start = sr_state_get_start_time_of_previous_protocol_run();
  } else {
    srv_start = sr_state_get_start_time_of_current_protocol_run();
  }

There is a bug here, because when we cross from the 23:00 consensus to the 01:00 consensus, the start of the SRV protocol changes and screws up the revision counter monotonicity.

This causes one descriptor batch upload to fail.