float inference for NNLF now broken

The current commit of VTM-11.0_nnvc is broken for float inference.

I checked out commit ea13b839 (VTM-11.0_nnvc). Then I changed line 61 in TypeDef.h to

using TypeSadlLFUnified=float;

Now I tried compressing using the old LOP3 model from the Sapporo meeting (VVCSoftware_VTM/models/nnlf_lop3_model_float.sadl).

This results in a crash when encoding:

started @ Tue Nov 19 21:58:18 2024 POC 0 LId: 0 TId: 0 ( IDR_N_LP, I-SLICE, QP 42 ) 16920 bits [Y 29.1805 dB U 34.7237 dB V 34.3962 dB] [ET 17 ] [L0] [L1] [MD5:9b4cb4e52ab044122148a062054f78e5,75788c914f8889a68839ee7464fad1ce,166b1f5a2a878cc4499fa6c1430bfe51] free(): invalid pointer

Doing the same for int16 inference works though.