fix(tunnels): normalize wing names in topic tunnel lookup for hyphenated dirs

[wahajahmed010]

Fix for #1194: Topic tunnels silently skipped for wings with hyphenated dir names

Root Cause

init stores wing names with hyphens/spaces replaced by underscores in mempalace.yaml (e.g. mempalace_public). But tunnel lookup functions (find_tunnels, list_tunnels, follow_tunnels) used the raw directory name (e.g. mempalace-public) as the lookup key, causing silent misses.

Fix

• Added _normalize_wing() helper that lowercases and replaces hyphens/spaces with underscores, aligning lookup keys with stored metadata
• Applied normalization in find_tunnels(), create_tunnel(), list_tunnels(), and follow_tunnels()
• Added logging.warning() calls when tunnel lookups return empty results for filtered queries — no more silent failures
• Backward compatible: single-token wing names (no hyphens) are unaffected

Testing

• 4 new tests in TestHyphenatedWingNormalization:
  • test_list_tunnels_filters_hyphenated_wing — verifies list_tunnels matches hyphenated input to stored underscore names
  • test_follow_tunnels_matches_hyphenated_wing — verifies follow_tunnels normalizes correctly
  • test_create_tunnel_normalizes_wing_names — verifies created tunnels use normalized names
  • test_find_tunnels_warns_on_empty_result — verifies warning log on empty results
• All 13 tests pass

[bensig]

Approve. Right shape for the lookup-side defense — small, focused, doesn't conflict with the init-side fix in #1195.

`pytest tests/test_palace_graph_tunnels.py` — 13 passed on this branch.

What landed

• `_normalize_wing()` helper in `palace_graph.py`
• `find_tunnels()` normalizes both query keys before lookup
• `create_tunnel()` normalizes both store keys on creation — important so all new writes go in normalized regardless of caller discipline
• Logger warning when a filter returns empty — actionable signal vs silent zero

One coordination ask before / after merge

#1195 (mine, on the init side) introduces `normalize_wing_name()` in `mempalace/config.py` as the canonical implementation. The two helpers are byte-identical (`name.lower().replace(" ", "").replace("-", "")`).

If both PRs land, the cleanup is one line: replace this PR's `_normalize_wing` with `from .config import normalize_wing_name`. Same logic, single source of truth. Either:

• Option A (preferred): rebase this PR after #1195 merges, swap to importing the helper. Minor touch-up.
• Option B: land both as-is; I'll open a one-commit follow-up that consolidates the two helpers.

Either works — flagging so we don't drift if someone later changes one normalizer and not the other.

Why both PRs should still land

The init-side fix prevents new bad keys from being written. The lookup-side fix here recovers gracefully from the bad keys that already exist in production palaces. Together they're complete defense; either one alone leaves a gap.

Ship it.

[wahajahmed010]

Thanks for the thorough review! Going with Option A — I'll rebase after #1195 merges and swap _normalize_wing for from .config import normalize_wing_name. Makes sense to have a single source of truth for the normalizer. Appreciate the coordination heads-up.

[igorls]

Rebased onto current develop (3474641) and pushed.

Conflict resolved in tests/test_palace_graph_tunnels.py — additive: kept this PR's TestHyphenatedWingNormalization class alongside develop's new TestTopicTunnels class from #1184. Both are independent and exercise different surfaces (lookup-side normalization vs. topic-tunnel computation), so they coexist with no logic interaction.

Note for follow-up (after #1195 merges): Igor's review comment on #1195 suggested folding this PR's _normalize_wing helper through from .config import normalize_wing_name once Ben's PR lands — single source of truth for the lower/replace rule across all four wing-slug producers. That swap is mechanical and a one-line change; can either land here as a follow-up commit after #1195 merges, or as a separate cleanup PR. Left as-is for now since normalize_wing_name doesn't exist on develop yet.

Tests: 1422 passed, 1 skipped on the rebased branch. ruff check + format clean. Author preserved on the original commit; committer is mine (force-pushed via maintainer-edit access).

Thanks for the lookup-side fix — together with #1195's producer-side normalization, this closes the hyphenated-wing tunnel loss for both new palaces (Ben's PR) and palaces with bad keys already on disk (this PR).