The restriction on matrix signals is not specific to the Biquad Filter, it is and Simulink HDL Coder wide restriction.
As each of the sections (M) of the biquad filters are chained together in series, you can achieve the same result by chaining together multiple (M) single section Biquad Filters, each of which will only require a vector input.
Note that each of the single stage Biquad filters would requires 2 scaling values to be specified, one scaling value at the input, and one at the output, meaning a input to g of vector size 2. This results in 2*M values of g specified for the full set of biquad filters. In your previous implementation input and output scaling values of the back-to-back sections are combined, hence why M+1 g values are required for the multi-section biquad filter implemented using a single block. These additional scaling values can be set to 1, i.e. the output g value for each Biquad Filter which is connected to the input of another Biquad Filter.
The disadvantage of this approach is that M cannot be configured at simulation run time.