Non-quantized QKeras layers break conversion

[vloncar]

Quick summary

While a bit nonsensical, QKeras layers without quantizers are legal. For example QDense(n_units) (without kernel_quantizer=something) fails conversion since we expect the quantizers to be present in the converter. In fact, the IR instance of Dense doesn't need a quantizer, so it is just the converter that is problematic.

Details

Steps to Reproduce

Create any model with QKeras layers and don't specify quantizers.

Expected behavior

The model should convert to IR.

Actual behavior

The converter fails.

Optional

Possible fix

We need to go through hls4ml/converters/keras/qkeras.py and check if supported layers pass without the quantizers specified. The check already exists for bias, we should add it for weights.

Additional context

I wish this was caught on QKeras side and creation of quantized models without a quantizer is prohibited, but that's out of our control.