Add abliteration built-in backend (Abliteration.ai)

[abliteration-ai]

Adds Abliteration.ai as a built-in backend.

Abliteration.ai is an OpenAI-compatible API hosting less-restrictive models (abliterated-model, abliterated-model-large) aimed at security research, which makes it a natural fit for an offensive-security harness. Docs: https://docs.abliteration.ai.

How it works

• New abliteration entry in builtinBackends, resolving its credential from ABLIT_KEY like every other provider's conventional variable.
• Unlike the existing built-ins, the engine does not know this provider's endpoint, so the built-in table gains an optional baseURL field and seeds https://api.abliteration.ai/v1 onto the backend (an explicit base_url in config still wins). Every other built-in is untouched - their endpoints stay in the engine.
• Provider resolution follows the existing convention: the backend name is the provider, and the engine accepts any provider once a base URL is given.

Usage:

export ABLIT_KEY="sk-..."
rook --backend abliteration --model abliterated-model "..."

Docs: README backend table + example, configs/rook.example.yaml, .env.example, desktop image PROVIDER_ENVS pass-through, CHANGELOG.

Tests: pinned built-in set updated; new TestAbliterationSeedsItsEndpoint covers endpoint seeding, credential resolution, and the base_url override. go build ./..., go vet ./... (both build variants) and go test ./... -count=1 -race all pass.