Add representative MoE timing experiment runners

[Happymic]

What This PR Does

This PR adds the P3 orchestration layer for representative MoE timing runs. It organizes the longer-running experiment drivers separately from the P2 pilot trace replay code.

Concrete Changes

• Adds window1_p3/run_p3_rev_subset.py for representative subset runs.
• Adds window1_p3/run_p3_scale_batch.py for scale/batch sweeps.
• Adds window1_p3/wait_p3_rev_stage_b.py to wait for and summarize Stage-B progress.
• Adds window1_p3/watch_p3_rev_raw_build.py to watch raw-build progress.
• Keeps P3 scripts in a dedicated window1_p3 package instead of mixing them into window1_p2.
• Removes old window1_p2.run_p3_* self-references and local hardcoded cwd defaults.

What This PR Does Not Do

• It does not change emulator timing logic.
• It does not commit experiment outputs, traces, logs, or model weights.
• It is orchestration for representative runs, stacked after the Qwen replay harness in Add Qwen3 MoE route-trace generation and replay #95.

Validation

Ran locally:

export PLENA_ROOT=/path/to/plena
source "$PLENA_ROOT/venvs/plena-py311/bin/activate"
export PYTHONPATH="$PWD:$PWD/PLENA_Compiler:$PWD/PLENA_Tools:$PWD/transactional_emulator/testbench${PYTHONPATH:+:$PYTHONPATH}"
python -m compileall -q transactional_emulator/testbench/window1_p3
python -m transactional_emulator.testbench.window1_p3.run_p3_rev_subset --help
python -m transactional_emulator.testbench.window1_p3.run_p3_scale_batch --help
python -m transactional_emulator.testbench.window1_p3.wait_p3_rev_stage_b --help
python -m transactional_emulator.testbench.window1_p3.watch_p3_rev_raw_build --help

After rebasing onto the portable P2 branch:

python -m compileall -q transactional_emulator/testbench/window1_p3
python -m transactional_emulator.testbench.window1_p3.run_p3_rev_subset --help