global state contamination issue in maptr loader cfgs

[kamalrajkannan78]

Issue

• Below issues were observed while running the test for variants present maptr_loader combinely not separately.
  • variants - TINY_R50_24E_AV2, TINY_R50_24E
  • exp1 : run av2 variant as last case - blank output oct3_z_maptr_av2_case_1.log
  • exp2 : run av2 variant as first case - shape mismatch related to num cams in second case. oct3_z_maptr_av2_case_2.log
• exp2 indicates that num_cams remains same across the tests, but it should be 7 for TINY_R50_24E_AV2 & 6 for TINY_R50_24E & exp 1 indicates there might be a global cfg contamination

Root cause

• This is due to the reuse of the same global config dict - gkt_cfg.
• In the loader I did attn_cfgs[1] = gkt_cfg (no copy). Then for AV2 I set attn_cfgs[1]["num_cams"] = 7, which changes the shared gkt_cfg itself.
• The next test gets that already-modified gkt_cfg, so num_cams sticks to the previous run’s value.

Fix

• deepcopy the gkt_cfg before mutation. do same for rest of the module-level cfgs.
  • oct3_z_maptr_av2_case_1_fix.log
  • oct3_z_maptr_av2_case_2_fix.log