Add deterministic DQN tests, CI checks, and local typecheck/test commands

[sbischoff-ai]

Motivation

• Replace a placeholder test with concrete, deterministic unit and smoke tests that validate core DQN components (replay buffer, epsilon-greedy action selection, target calculation and learning-step invariants).
• Make tests deterministic using fixed RNG seeds so CI and local runs are reproducible.
• Ensure CI and local developer workflows explicitly run lint, type checks (including tests), and unit tests to catch regressions earlier.

Description

• Replaced tests/dummy_test.py with deterministic tests that cover ReplayBuffer, EpsilonScheduler / epsilon-greedy behavior, DQN target/loss calculation and learning-step invariants (global_step, epsilon update), and a short smoke integration run on CounterEnv (seeds are set via a helper _set_all_seeds).
• Updated CI workflow .github/workflows/ci.yml to run explicit quality commands: uv run black --check adeptly tests, uv run mypy adeptly tests, and uv run pytest -q.
• Updated Makefile typecheck target to mypy adeptly tests so local make typecheck checks tests as well as package code.
• Updated README.md development/checks section to document the same direct commands for local validation (uv run black --check adeptly tests, uv run mypy adeptly tests, uv run pytest -q).

Testing

• Ran uv run black --check adeptly tests and it completed successfully (no formatting changes reported).
• Ran uv run mypy adeptly tests and it reported Success: no issues found.
• Ran uv run pytest -q and all tests passed (15 passed), verifying the new deterministic tests and existing suite.

---

Codex Task