Fix bugs, including PBCDataloader incorrect preprocessing

[DrShushen]

Description

Fix three bugs:

1. 1e3d6b4 fixes the "did not converge" error in tests involving XGBSurvivalAnalysis/XGBTimeSeriesSurvival. The tests that would previously (at least in some cases) fail include tests/metrics/test_performance.py::test_evaluate_performance_survival_analysis[PerformanceEvaluatorXGB-test_plugin0]. The fix was to change the strategy from "weibull" to the more stable "debiased_bce".
2. f309351 fixes the failing check in WindowLinearLayer.forward, which would fail when a length zero tensor was provided (i.e. no static data case). The conditional only needs to run if static data is expected, so an extra check added to fix this. The tests that would previously (at least in some cases) fail include: tests/plugins/core/models/test_ts_tabular_vae.py::test_ts_vae_generation[GoogleStocksDataloader].
3. 1e3d6b4 fixes incorrect preprocessing in PBCDataloader.
   • Steps to replicate:
     • If you add this sanity check, which checks that the static features extracted from the source data are indeed unchanging (over time steps) for each patient...
     • ...You find that this fails in the original implementation (run from synthcity.utils.datasets.time_series.pbc import PBCDataloader; PBCDataloader(as_numpy=True).load()).
   • Why is this happening? The dataframes used in preprocessing steps mix up the indexes of the samples at the sorting step, and at a later step attempt to reference the old indexes - which leads to mixing up different patients' data together. This is now fixed by resetting the index at the sorting step.

Affected Dependencies

None.

How has this been tested?

• No new tests are required.

Checklist

• I have followed the Contribution Guidelines and Code of Conduct
• I have commented my code following the van der Schaar Lab Styleguide
• I have labelled this PR with the relevant Type labels
• My changes are covered by tests

[DrShushen]

@bcebere for 3. above, are there any performance tests that we expect to be affected? Since PBCDataloader now essentially returns different data.

[DrShushen]

@bcebere
Getting this following error in GH actions but not locally
https://github.com/vanderschaarlab/synthcity/actions/runs/4430650800/jobs/7772687428

2023-03-15T21:05:55.6778011Z FAILED tests/plugins/privacy/test_decaf.py::test_plugin_generate[test_plugin0] - RuntimeError: Training with multiple optimizers is only supported with manual optimization. Set `self.automatic_optimization = False`, then access your optimizers in `training_step` with `opt1, opt2, ... = self.optimizers()`.

Looks like possibly a pytorch-lightning v2 issue.