Support for custom science Workflows

[phycodurus]

This is just an idea to support the defintion and execution of custom workflows. For example, if a workflow consist of steps such as:

1. receive alert
2. create target (or region)
3. perhaps identify targets of interest with that region
4. do analysis of target(s) perhaps to categorize, prioritize, or determine cadence of follow-up observations
5. make observations
6. analysis of follow-up observations
7. loop back to step 3 or ???

The idea is to provide a frame work with integration points where by a (perhaps yaml-defined) workflow could be started and tracked (perhaps with a Jenkins-like UI).

Again, this is just an idea.

[phycodurus]

Latest thinking on this is incorporating prefect with adapters that let TOM developers focus on writing @tasks and @flows without having to worry too much about the infrastructure.
Here's an idea:

• Use Prefect to check for outdated time series gulfofmaine/buoy_barn#309

@KKruszynska, we'd like to hear about your MOP use case, so we can provide support. Please feel free to comment.
@griffin-h, your SAGUARO thoughts are welcome.
@Fingel, fyi

[KKruszynska]

Hi, here is an example of a workflow we would like to implement into MOP at some point (to some extent it's already there).

A series of targets receive new data (from surveys, follow-up, etc.). For each, we would like to find new best-fitting models and evaluate if there are any ongoing anomalies (meaning, the presence of a planet or a second object in the lens). At later steps, this information would be used to assign the correct observing strategy. Ideally, this would have been done in parallel and there would be a controller process that assigns each free thread a new target.

A flow for a single target would look like this:

1. Perform fitting in the "simple" model category (point source point lens microlensing model (PSPL) with or without blending and/or parallax effect).
2. Evaluate found models and select the best-fitting one.
3. Check how many points are in the light curve. If there is a sufficient number...
4. Attempt an "intermediate" model fit (PSPL and finite source effect, with or without parallax effect).
5. Evaluate new-found models with the old ones and select the best.
6. Perform a check for anomalies. If an anomaly is found...
7. Perform fitting of "advanced" models (binary model, with or without parallax and/or finite source effect)
8. Evaluate new models and select the best fitting one.
9. Perform a check for anomalies in the results that were selected for an advanced model. Report the outcome of the check.

A result of such flow would be a best-fitting model assigned to the event, along with the results for all found models, some statistics used to evaluate the quality of the fit and its significance and the results of the anomaly checks.

Let me know if that makes sense. @rachel3834 feel free to add or remove anything from the list I wrote above.