Fix: Simplify prompt caching for new Anthropic API

[enyst]

Simplify prompt caching for new Anthropic API by marking only the last user or tool message as cacheable. Updated unit tests accordingly.

Reference: Anthropic simplified their implementation a little, and we can too.

---

To run this PR locally, use the following command:

docker run -it --rm   -p 3000:3000   -v /var/run/docker.sock:/var/run/docker.sock   --add-host host.docker.internal:host-gateway   -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:efc368e-nikolaik   --name openhands-app-efc368e   docker.all-hands.dev/all-hands-ai/openhands:efc368e

[enyst]

I can confirm testing locally, that the reported cache hits/writes look as expected:

... step 21
Input tokens: 12572 | Output tokens: 103
Input tokens (cache hit): 12331
Input tokens (cache write): 238

... step 22
Input tokens: 12699 | Output tokens: 53
Input tokens (cache hit): 12569
Input tokens (cache write): 127