feat(logging): name the offending weight when compression is skipped

[pkmandke]

When a weight can't be compressed because its shape is incompatible with the block/group size, the skip warning now names the tensor and its shape. Previously quantization logged only the compression-target enum, and palettization logged no identifier.

Quantization:

# before:
Tensor (target: CompressionTargetTensor.WEIGHT) incompatible with block size configuration: Tensor size 12 along axis 0 is not divisible by block size 8. Skipping quantization.

# after:
Tensor '0.weight' (target: CompressionTargetTensor.WEIGHT, shape: (12, 8)) incompatible with block size configuration: Tensor size 12 along axis 0 is not divisible by block size 8. Skipping quantization.

Palettization:

# before:
Tensor incompatible with granularity: Tensor size 12 along axis 0 is not divisible by group_size 8. ... Skipping palettization.

# after:
Tensor '0.weight' (shape: (12, 8)) incompatible with granularity: Tensor size 12 along axis 0 is not divisible by group_size 8. ... Skipping palettization.

Implementation Detail:

A module can't discover its own FQN from inside forward(), which is where we actually log the warning. So added a tensor_fqn attribute to CompressionSimulatorBase, and populated it during prepare() (before the first forward):

• eager: walk parametrizations and for every simulator in a ParametrizationList capture the parameter name it compresses.
• graph: read the FQN off the get_attr node feeding each simulator.

Note: This is for weights only.