Review follow-up: changes required on PR #84 #86

Required changes from review of PR #84 (recursion fix in src/trends.rs):

• Major — spike/outlier split points are silently dropped. IEEE-754 simulation confirmed: spike series produces segments [0,24] + [26,49], index 25 is missing, contradicting the PR's validation claim.
• Major — recursion depth remains O(n) worst-case (n=10k convex series -> depth ~4052), so large inputs can still overflow the stack.
• Minor — 2-element input with threshold=0.0 returns zero segments.
• Minor — no new regression tests; existing test only checks len >= 2.

Please address before merge.