Perf/vectorize previous high low: ~80-100x speedup

[Favian0363]

Vectorizes previous_high_low() method by replacing the per-candle loop

Problem

The current implementation loops through every candle and calls np.where() to find resampled periods

This creates O(n × m) complexity where n = candles, m = resampled periods.

Solution

Use np.searchsorted for O(log m) binary search
Eliminate loop completely
Use groupby().cummax & cummin for vectorized broken flag calc

Benchmarks

Before vectorization:
..previous_high_low test time: 1.3034696578979492
previous_high_low test time: 1.2444789409637451
previous_high_low test time: 1.194727897644043

After vectorization:
..previous_high_low test time: 0.010783195495605469
previous_high_low test time: 0.01098489761352539
previous_high_low test time: 0.00976705551147461