Decide: does --cross-project let global learnings outrank local ones? #61

Genuine behavioral decision, not a bug to auto-fix.

bin/idstack-learnings-search:70 slices matches[-limit:] while sources are ordered local-then-global. Under --cross-project, that means global entries win when the limit truncates — which contradicts the comment directly above at :69 claiming local takes precedence.

One of the two is wrong. Either:

• the comment is right and the slice should be matches[:limit] (local wins), or
• the behavior is right and the comment should go.

Leaning toward local-wins: a project-specific learning is more likely to be the relevant one, and that's what the comment says the author intended. But it's your call — this changes what designers see.