1. Forced hash collisions are fragile in comparison to the genuine material. CSAM hashing is devised in a way to allow change in the image without significant variation to the hash. Forced collisions however don't possess that property as their true hash is different.
2. It's trivial to detect images with false hashes, and different hashing methods will not produce collisions.
3. The task you're proposing, in itself is not pragmatic nor effective: dissenting political materials are easily created fresh (i.e. no colliding hash) and exists in many forms that go well beyond an image.
4. Human reviewers again would see the pattern and adapt the system accordingly
5. CSAM database holders aren't idiots. They understand the significance of their hash libraries and work against manipulation from CSAM peddlers.