feat(math): classify resonances high-Z/low-Z; avoid only high-Z

[dc0sk]

Summary

Adds the physically-nuanced resonance handling deferred from #127. A wire is resonant at every quarter-wave multiple, but the feedpoint impedance alternates:

• low-Z — odd multiples (λ/4, 3λ/4, …), current-fed, ~35–50 Ω, near 50 Ω and easy to match;
• high-Z — even multiples (half-wave multiples λ/2, λ, …), voltage-fed, hundreds–thousands of Ω, genuinely hard to match.

The non-resonant optimizer now avoids only the high-Z resonances, leaving the desirable low-Z lengths available, and every resonant point is tagged in the display and exports so the recommendation stays reconcilable against the list.

Why this is the consistent version of the earlier idea

#127 deliberately did not split resonances into high/low-Z, because doing so partially would create a second, conflicting definition of "resonance" versus the display/export. This PR does it fully: the tags appear everywhere, so recommending a length near a low-Z point (e.g. 30.15 m near the low-Z 3× = 30.14 m) is self-explanatory rather than contradictory.

Changes

• new ImpedanceClass (Low/High) from harmonic parity
• build_non_resonant_resonance_points keeps only even (half-wave) harmonics, padded by one half-wave per band so near-edge clearance stays honest and the set is never spuriously empty
• ResonantPoint gains impedance_class; display lines are tagged [low-Z]/[high-Z] with a legend; JSON export gains "impedance_class"; text/CSV points tagged inline
• docs/math.md §5 updated

Test plan

• impedance_class_follows_harmonic_parity, non_resonant_avoids_only_high_z_resonances (asserts even harmonics avoided, odd left available, first beyond-edge high-Z included), resonant_points_display_lines_tag_impedance_class_and_legend
• cargo test all pass (370 lib + integration), clippy + fmt clean
• Verified: non-resonant 40m now recommends near the low-Z 30.14 m point with honest 33% clearance vs the high-Z points

🤖 Generated with Claude Code

https://claude.ai/code/session_01Jf8uBU8AYxkrG8fa8yo6tC

[dc0sk]

Applied the high-effort review findings:

• [1][3] Relabeled the non-resonant clearance as high-Z clearance (it measures distance to the hard-to-match high-Z resonances) in the results view and advise notes, with a one-line explanation — a recommendation may intentionally sit on an easy-match low-Z resonance, so the generic 'resonance clearance' label was misleading.
• [2] Reverted the resonant_points_in_window CSV/text value to the stable {h}x={len}{unit} contract format; the class stays a first-class JSON field and is shown tagged in the results view. Added a contract-guard test.
• [4] Fixed stale docs ("within 1 m outside" → one half-wave pad).
• [5] Moved the ImpedanceClass enum out from between band_resonant_points_m's doc and its fn.

All 461 tests pass; clippy + fmt clean.