docs: update README and add architecture & migration guides

[sbischoff-ai]

Motivation

• Bring repository documentation up-to-date with the current PyTorch-first implementation and clarify the migration away from TensorFlow 1.x.
• Provide a clear architecture overview for agent components, data flow, and the multimodal observation pipeline so maintainers and users can understand core subsystems.
• Give concrete migration examples for users updating from legacy DQNAgent imports and AdeptlyEngine usage to the new API surface.
• Avoid committing regenerated site artifacts while still documenting how to rebuild the generated /docs output locally.

Description

• Updated README.md to reflect current capabilities, roadmap priorities, explicit TensorFlow 1.x migration messaging, and a note explaining how to rebuild generated docs with uv run make docs without rebuilding /docs in this change.
• Added ARCHITECTURE.md which documents agent components, DQN/trainer data flow, the real-time actor/learner split, and the multimodal observation pipeline (schema, normalization, encoders, and fusion).
• Added MIGRATION.md which provides before/after code examples and guidance to migrate from legacy adeptly.dqn.DQNAgent and AdeptlyEngine patterns to adeptly.agents.dqn.DQNAgent, DQNTrainer, and RealTimeInferenceLoop.
• Kept changes scoped to documentation only and preserved existing /docs generated artifacts; no generated site files were rebuilt or committed as part of this PR.

Testing

• Ran static type checks with uv run mypy adeptly tests which reported no issues.
• Ran the test suite with uv run pytest -q, which initially failed during collection due to missing runtime dev dependencies (e.g. numpy, torch).
• Synced the development environment with uv sync --frozen --extra dev to install required packages and re-ran uv run pytest -q, resulting in all tests passing (15 passed).
• No production code changes were made, and the existing unit tests passed after environment sync.

---

Codex Task