Further improvements to `abnormal_lab` tasks.

Each of the abnormal lab tasks has a very similar structure right now, merely differing in which abnormal lab is of interest. Namely, they all have the following properties:

1. The input is the first 24 hours of a hospital admission.
2. They all guarantee that the patient neither dies nor is discharged before 48 hours after hospital admission.
3. The prediction target is the presence of an abnormal lab between 24 and 48 hours of the hospital stay.
4. They all exclude children.
5. They all exclude those who have shown any abnormal lab results of the target lab any time prior to the end of the input window.
6. They are all testing conditional observation of an abnormal lab -- meaning that they are testing whether an abnormal lab will be observed, conditioned on any measurement of that lab being observed.

There are several problems I see with these tasks in light of these commonalities:

1. MIMIC-IV does not have general hospital data -- rather, it only has hospitalization records for patients who have or will eventually be seen at the BIDMC ED or ICU (see #153). This means that these tasks, as stated, are actually not appropriate for MIMIC-IV!
2. The restriction that the patient never has an instance of these abnormal lab results anytime in their historical record seems very conservative. Most patients who are seen in an ICU will have some recorded instances of abnormal labs, I imagine.
3. They lack censoring protection -- while this is not a universal desiderata (e.g., see #9 and #10) it is potentially a good thing to have to ensure you have well-founded records included.
4. Is the right age range here really just excluding children? Or is a more targeted age range more clinically meaningful?
5. All of these labs use universal ranges for men and women -- I'm not sure if that is medically appropriate (though it is very common, so I suspect this is fine for now).
6. The names used are not super clear, as they don't reflect in particular the conditional aspect of these tasks.
7. I like the conditional aspect of this prediction, as we don't want this task to really just become "lab prediction" in a manner that is unclear to users; however, I'm uncertain if it is the right way to accomplish its goal. Namely, I think what we want is to track, for patients for whom lab X is a concern and is being tracked but so far has been stable, whether or not it will become unstable within the next 24 hours. That is a bit different than what we're tracking so far. Perhaps a better way to do this conditionality is to either add (and keep the conditionality on measurement) or replace the conditionality on measurement condition with a condition that a normal observation of the lab in question is observed in the ICU/hospital stay to date? That way we are only examining patients who are being monitored for this lab? We could also consider adding a task that is "will this lab be measured at all" as a task to "complete" the probabilistic landscape here.

I'd like to suggest the following steps in light of these issues:

1. We consult with clinician experts on the suitability of these tasks, both generally and in light of the suggested alterations below. (in particular, @Gallifantjack and @mandiehyewon, please feel free to chime-in with any inputs, particularly on questions about age ranges, gender-specific value cutoffs, and the conditional aspect of this prediction).
2. We shift these tasks to ICU tasks, not hospital tasks. They would then be suitable for MIMIC-IV, eICU, AUMCdb, NWICU, etc.
3. Given both the general concern and, more specifically, the shift to ICU, we change the criteria of "never having seen an abnormal lab result of this type" to "never having seen an abnormal lab result of this type within this hospitalization". This will help focus this task towards identifying patient decompensation in ways that are novel within the scope of this hospitalization.
4. We add a simple censoring protection window (asserting that there are any measurements observed after the end of the target window).
5. We rename these tasks, with a new naming scheme to be suggested in the comments of this issue.
6. We make a possible alteration to the conditional aspect of the scheme concordant with expert advice, e.g., by adding a criteria that the lab is already being monitored closely for this patient.