Decompose intermittent-power execution time into exec, checkpoint, and wait

[byeongjee]

Summary

• Pulse P3.5 around each capacitor wait so the Saleae's second channel measures wait time (documented in docs/intermittent.md and docs/saleae.md); a death during a wait extends the span to the final exit.
• Extract wait spans in ckpt intermittent and record wait count, deaths, and wait time in the summary CSV.
• Add -D to bench commands and build the continuous-power reference with INTERMITTENT_BUILD (the same wait-mode binary run under continuous 3.3 V).
• Charge a 64-byte upper bound for memcpy/memset with non-constant size in bb-energy-analyzer; re-measure sha256 and poly1305 accordingly.
• results/intermittent/summarize.py decomposes T_total into T_exec, T_ckpt, T_wait, and the remainder, and emits the aggregate and per-benchmark LaTeX tables used in Section 6.3.

Test plan

• uv run pytest tests/test_intermittent.py
• Re-ran the intermittent-power matrix and regenerated plots and tables