Pushed onto main.

I fixed the problem on Windows. Malte, do you want to review that issue or should I proceed to draft a commit message?

Experiments finished and look ok (no large changes; the small ones probably due to reaching time limits): https://ai.dmi.unibas.ch/_experiments/ai/downward/issue1019/data/issue1019-v1-eval/issue1019-v1-issue1019-base-issue1019-v1-compare.html

I restarted this and planned to break the changes up into several small chunks. This issue only moves everything related to computing a PatternDatabase into a new class PatternDatabaseFactory (plus an external function as abbreviation).

Pull request: https://github.com/aibasel/downward/pull/144

I took the changes which I experimentally made while working on issue1007 and created pull request: https://github.com/aibasel/downward/pull/40/files
This is not very polished yet but should serve as a starting point for discussing the design further.

Currently, PatternDatabase objects are created from their constructor. This involves computing hash multipliers for hashing states, abstract operators and a match tree for successor (or rather predecessor) generation, and distances for the actual PDB. Separating these steps into individual functions/factory classes has the benefit that one can use these parts independently of computing a PDB. It further allows to hook into the PDB computation to compute any additional information as a side product, without affect the PDB interface. issue1007 shows a use case for this scenario, and two other projects I've recently worked on would benefit from this.