Implement Further Feedback on README files

[oleschwen]

The README files for the different roles contain todos and should generally be improved based on the feedback from the first ODELIA-wide swarm training using MediSwarm. First feedback was merged as part of #96, continuing in a new issue.

[oleschwen]

For future tests, we should parse the standard annotations table rather than extracted .csv tables (as we did now for compatibility with the Challenge). For this purpose, we may want to extend the preflight check to verify that the more numerous columns are parsed correctly.

[oleschwen]

We may also want to print more meaningful error messages if files are not found or do not contain the correct information. However, as training logs might be shared, we should not leak information about private datasets when doing so.
One approach could be to explicitly check if the required metadata tables are present and to report summary numbers of the number of images listed in the table, number of image files present, numbers of image in one but not the other.
If we want to log identifiers in the tables/present as files and possibly hashes of the files (which would be useful for training documentation purposes, but that is a separate feature), this should go in a separate log that is not to be shared.

[oleschwen]

Not for the README: we could make num_workers optionally adaptable in startup/docker.sh.

[SteffenRenisch]

• add a hint to take care that there is only one VPN connection per partner open.
• clarify location of data folder

[SteffenRenisch]

Add to "troubleshooting" section:

• recommend some network testing (e.g. repeatedly pinging the VPN server for an hour or so - maybe think about better test scenario) - Goodaccess noticed packet loss every couple of minutes in some occasions.