CI: Add discrepancy testing for halld_recon

[nathanwbrei]

In JANA1, users use the janadump or hd_dump plugins to inspect data event-by-event. Right now, we are manually using hd_dump for discrepancy testing our halld_recon port.

In JANA2, we have several different tools that capture different pieces of the necessary functionality but definitely need reorganization:

• The ported hd_dump
• JInspector
• regressiontest
• JCsvWriter

We also should consider whether we can simply reuse ePIC's capybara tool for detecting discrepancies instead.

Requirements:

• Experiment-agnostic. Leverages JObject::Summarize to incorporate users' preferred formatting and also be consistent with janaview.
• Accessible from inside a debugger (without restarting the process)
• Accessible when running in interactive mode (i.e. JApplicationInspector)
• Outputs ASCII tables, pretty-printed JSON, or CSV, depending on user config
• Outputs to a separate file, not to stdout, so that other log messages don't pollute our diff
• Works with topologies that use multiple event levels (e.g. TimesliceExample).
• Sorts outputs by event number so that output is deterministic even when multithreading is enabled

Nice-to-haves:

• Works with PODIO data (without needing to know about the exact PODIO datamodel)

Obvious problems with the existing tools:

• JCsvWriter only works on one collection at a time and can't be made into its own plugin due to how it uses templates. This is an easy fix, although JCsvWriter should probably be outright replaced by a subset of JInspector anyhow.
• JEventSource implements nevents/nskip in a way which makes all of these tools very inefficient. Adding random file access and O(1) nskip will fix this.
• regressiontest attempts to do a diff itself, which it probably shouldn't