Hash Function for Tile Coding

Implement a tile coding function that maps continuous state variables to a set of active tile indices using a hash function. Tile coding is a method for discretizing continuous spaces used in reinforcement learning function approximation. Your function should: 1. For each tiling (0 to num tilings 1), apply a displacement offset to the state before discretization. The offset for tiling t is t / num tilings (in tile width units). 2. Normalize each state dimension to [0, tiles per dim] using the provided bounds, then add the tiling offset. 3. Discretize each shifted dimension using floor to get integer tile coordinates. 4. Construct a coordinate tuple: (tiling index, tile coord dim0, tile coord dim1, ...). 5. Hash this coordinate tuple into an index in the range [0, memory size) using a djb2 style hash function. 6. Return a list of active tile indices, one per tiling. The djb2 hash works as follows: initialize h = 5381, then for each integer c in the coordinate tuple compute h = ((h 33) + c) masked to 32 bits. The final index is h modulo memory size. Args: state: list of floats, continuous state variables num tilings: int, number of overlapping tilings tiles per dim: int, number of tiles per dimension per tiling memory size: int, size of the hash table (weight vector length) state bounds: optional list of (low, high) tuples per dimension; defaults to (0.0, 1.0) for each dimension Return a list of integers representing the active tile indices.