Add integration to upload results to some "pastebin"

[yarikoptic]

We should not implement those integrations but use some external library/toolkit for that. But it should be optional (configurable). I personally would prefer helper which could be configured to use https://paste.debian.net/ .

• Might also need to be configured to upload only last N lines etc.
• May be there is a service which could get .md and with figures so we could even configure to upload rendering of _stats.json progression
• _info.json better be uploaded pretty-printed

[yarikoptic]

I think we should look into some simple setup for

• uploading results to a possible wide range of storage solutions. Most of them would return back some kind of ID or UID or URI/URL to identify that submission
• During EAB meeting we also contemplated uploading of "results" into local ReproLake's (graph DB) so they could potentially later queried etc ("did I see this error before") etc
• @jwodder uses duct to collect operational stats on s3invsync and post them into a github issue where he collects samples Run Reports dandi/s3invsync#106 . I could see potentially it being github wiki etc. But interesting aspect there -- desire for markdown rendering of the summary
• Different "storage backends" would have potentially differing desires for what is to be stored between collected stdout/err, stats, and info.
  • Quite often full outputs would need to be either
    • filtered (e.g. using regex as simplest way)
    • truncated
      • last X lines
      • lines collected in the last Y seconds of running (would require idea: --timestamp-outputs or --json-outputs #213)
• Provided information might want to be extended with additional information relevant to the use case so to help later reproduction or determining context
  • e.g. if used within some datalad run -- we might want to accompany with output of datalad wtf
  • I was considering integrating with a more generic https://github.com/sosreport/sos but it might be more oriented toward sysadmins / heavier debugging (ref)
• It would be up for the "plugin" to configure sensible defaults, so for user it should be really easy to get started
• Some backends might immediately provide response/recommendation given a specific error to troubleshoot or potentially even request providing additional information. This way we could hook in some LLM-based solutions
  • although sounds like a separate development, it would actually first depend on ability to "upload"! hence worth noting that additional interface/interactions might be desired in some "upload" and there might be need for interaction (even if just to set credentials)