[REF] Refactor anomaly_detection into submodules

[MatthewMiddlehurst]

Describe the issue

Currently, all anomaly_detection estimators are in the top level module directory. We should separate them into submodules. The current proposal found through discussion in module meetings is separating by algorithm family (see https://dl.acm.org/doi/abs/10.14778/3538598.3538602 by @SebastianSchmidl, Table 1).

Including a few extras, this would give us:

• "forecasting"
• "reconstruction"
• "encoding"
• "distance"
• "distribution"
• "isolation_tree" (too specific? Maybe "outlier_detection" or just "trees"?)
• "whole_series"/"collection" (See [ENH] Base class for whole-series (collection) anomaly detection #2662)
• "deep_learning"

Suggest a potential alternative/fix

Categorise current estimators and refactor the code in the module.

Additional context

Discussed in an anomaly detection module meeting. https://hackmd.io/@aeon-toolkit/HyZVExBZ1l

[Kaustbh]

@aeon-actions-bot assign @Kaustbh

[Kaustbh]

@KowshikB03 Thanks for your effort in addressing this issue. However, after reviewing your PR, I noticed that you created single estimator subfolder and moved all algorithms there. The intention, as mentioned in the issue, is to create different submodules/subfolders based on the categories outlined in the referenced paper.

@MatthewMiddlehurst I have categorized the estimators according to the suggested submodules. May I go ahead and create a PR?