[Shared Utils & Models] Add basic uniform replay buffer & checkpoint serialization

[saviornt]

Description

Implement a simple uniform/circular replay buffer and basic checkpoint save/load for Q-table + metadata. This enables experience sharing between edge episodes and central training.

Why: Replay is essential for off-policy learning; checkpoints allow policy transfer edge ↔ central.

Type

• Task

Focus Area (pick one)

• Shared Utils & Models

Priority

• High

Acceptance Criteria

• UniformReplay class (append, sample(batch_size), len, max_size control)
• Checkpoint serialization: save/load Q-table (numpy .npz) + config + episode stats
• Save → load round-trip preserves Q-values (float32 tolerance)
• Tests: buffer overflow behavior, sampling uniformity, checkpoint integrity
• Google-style docstrings on public API
• Located in shared/src/learning/q_learning/replay/ and io/serialization.py

Blocker / Dependencies

• [Shared Utils & Models] Create Q-Learning config & types (Pydantic v2)

Notes / Links

• Future extension: prioritized replay in same module