Regression in 2025.w18: 32 HARQ processes show re-transmissions at OAI UE

Enabling 32 HARQ processes, e.g. with the following change makes the UE show re-transmissions, even though the gNB does not indicate re-transmissions at all.

diff --git a/ci-scripts/conf_files/gnb.sa.band78.106prb.rfsim.conf b/ci-scripts/conf_files/gnb.sa.band78.106prb.rfsim.conf
index 9627a48d87..a6a294c327 100644
--- a/ci-scripts/conf_files/gnb.sa.band78.106prb.rfsim.conf
+++ b/ci-scripts/conf_files/gnb.sa.band78.106prb.rfsim.conf
@@ -24,6 +24,8 @@ gNBs =
     ////////// Physical parameters:
 
     min_rxtxtime = 6;
+    num_dlharq   = 32;
+    num_ulharq   = 32;
 
     servingCellConfigCommon = (
     {

This can be tested by running the gNB and UE with these commands:

sudo ./ran_build/build/nr-softmodem -O ../ci-scripts/conf_files/gnb.sa.band78.106prb.rfsim.conf --rfsim

sudo ./ran_build/build/nr-uesoftmodem -O ../ci-scripts/conf_files/nrue.uicc.conf --band 78 -C 3319680000 -r 106 --numerology 1 --ssb 516 --rfsim

The output at gNB side looks OK:

[NR_MAC]   Frame.Slot 0.0
UE RNTI 4287 CU-UE-ID 1 in-sync PH 48 dB PCMAX 20 dBm, average RSRP -44 (16 meas)
UE 4287: dlsch_rounds 40/0/0/0, dlsch_errors 0, pucch0_DTX 0, BLER 0.00182 MCS (0) 0
UE 4287: ulsch_rounds 543/0/0/0, ulsch_errors 0, ulsch_DTX 0, BLER 0.00000 MCS (0) 0 (Qm 2 deltaMCS 0 dB) NPRB 5  SNR 50.5 dB
UE 4287: MAC:    TX            663 RX           7261 bytes
UE 4287: LCID 1: TX              6 RX            122 bytes

But the UE shows these logs:

[NR_MAC]   UE 0 stats sfn: 896.8, cumulated bad DCI 0
    DL harq: 24/15
    Ul harq: 273/258 avg code rate 0.1, avg bit/symbol 2.2, avg per TB: (nb RBs 5.4, nb symbols 9.6)
[NR_MAC]   UE 0 stats sfn: 0.8, cumulated bad DCI 0
    DL harq: 25/15
    Ul harq: 273/271 avg code rate 0.1, avg bit/symbol 2.2, avg per TB: (nb RBs 5.4, nb symbols 9.7)

I verified that this issue is not present in 2025.w17.