Fixed:

- CompletionPolicy::Discard was not working as expected, only in case of forwarding. This was apparently the only case we had so far.

    /// Do not run the ProcessCallback. Contents of the record will
    /// be forwarded to the next consumer, if any.
    Discard,

Therefore devices which were expecting sporadic input could enter the run method although the input was not available, leading to seg faults. Fixed by https://github.com/AliceO2Group/AliceO2/pull/12578

- Some devices need to run in a mode where they receive only CCDB objects and no other input data in order to produce output needed by other devices
    - run() is avoided for devices only in case it has sporadic inputs, not if there are only TF and/or condition inputs
    - fixed by https://github.com/AliceO2Group/AliceO2/pull/12576
    
    
Pending:

• We still have spurious errors when running the FST: [1746295:ctf-writer]: [10:36:24][ERROR] Some Lifetime::Timeframe data got dropped starting at 1
• When running the FST with local aggregator and proxys the output proxys are dying without stack trace
• Lifetime::Timeframe data is dropped randomly (reproducible on staging, but not with a single TF) because of CCDB objects missing. Example:

[marten@login epn279]$ grep GLO/GRPMAGFIELD mch-track-finder_reco2_2024-01-23-19-03-16_12421396949842584864_out.log
[19:19:27][INFO] Caching in 22864921167504 ptr to GLO/GRPMAGFIELD/0 (0x5ffbae0)
[19:46:56][ERROR] Missing GLO/GRPMAGFIELD/0 (lifetime:condition) while dropping incomplete data in slot 1 with timestamp 16552 < 16553.