[Config] Add the missing dense-path keys to the heterogeneous config

[YWHyuk]

Problem

scripts/stonne_experiment/run.sh runs three STONNE experiments. The third one --
the heterogeneous run (test_hetro.py --mode 2, sparse core + dense core on two
partitions) -- currently dies as soon as the dense matmul kernel is launched:

File "PyTorchSimFrontend/extension_codecache.py", line 322, in run_kernel_simulation
  if extension_config.pytorchsim_functional_mode and not autotune:
File "PyTorchSimFrontend/extension_config.py", line 57, in __getattr__
  return config_yaml['pytorchsim_functional_mode']
KeyError: 'pytorchsim_functional_mode'

configs/heterogeneous_c2_simple_noc.yml is missing three keys that the dense
(ws_mesh) path requires: pytorchsim_functional_mode, pytorchsim_timing_mode
and core_spad_size_kb. This is config bit-rot -- the file was last touched by a
DRAM-config change and never picked up keys that later became mandatory. The
sparse-only (--mode 0) and dense-only (--mode 1) experiments were unaffected
because they use different config files.

Fix

Add the three keys. Values match
systolic_ws_128x128_c1_simple_noc_tpuv3_half.yml, the config the dense-only
experiment already uses. core_spad_size_kb: 16384 is consistent with the
128 lanes x 128 KB per lane this file already declares.

Verification

On develop (e671371), all three experiments from run.sh at M=N=K=1024,
sparsity 0.9:

mode	config	before	after
0 (sparse only)	stonne_big_c1_simple_noc	26696 cycles	unchanged
1 (dense only)	systolic_ws_128x128_..._tpuv3_half	81720 cycles	unchanged
2 (heterogeneous)	heterogeneous_c2_simple_noc	KeyError	48177 cycles

The standalone sparse path (tests/system/test_stonne.py) was also checked and
scales with sparsity as expected -- at size 32: 50990 / 28022 / 8488 cycles at
sparsity 0.0 / 0.4 / 0.8.

Note (not fixed here)

scripts/stonne_experiment/run_trace.sh greps for Stonne Core [0] : Total cycle
but the simulator now prints StonneCore [0] : Total cycle (no space), so that
script silently reports empty cycle counts. It also hardcodes /workspace/PyTorchSim
and runs the same script twice while labelling one row "TOG". Left alone since the
original intent is unclear.

🤖 Generated with Claude Code

https://claude.ai/code/session_01D3QZ8xomsrbs4SoA5wLQvJ