Changelog

`main`

PIP: replaced lazy-deletion PriorityQueue with a flat array-backed max segment tree, eliminating the O(n·k) stale-entry accumulation that caused 25× slowdown at large k
PIP now operates at true O(n + k log n): O(n) segment tree build + O(log n) per distance update; the heap's lack of decreaseKey is no longer a bottleneck
PIP 100k/k=1000: 7.75 s → 959 ms (8× faster); 100k/k=5000: 8.15 s → 1.01 s (8× faster)
PIP 500k now completes in 4–11 s (previously hung indefinitely)
PIP 100k/k=100 regressed from 304 ms to 631 ms — a known trade-off: the segment tree builds eagerly in O(n) regardless of k; the old heap built lazily and only ever touched O(k) entries when k ≪ n, so it was cheaper in that regime. The crossover is k/n ≈ 0.005–0.01: below that the heap was faster; above it the heap's stale-entry accumulation dominates. The segment tree wins decisively for k/n ≳ 0.01 (targetSize ≳ n/100)

DoublePoint is now a record — correct equals/hashCode/toString out of the box, and the JIT can inline record accessors more aggressively than regular getters
Point interface accessors renamed: getX() → x(), getY() → y(), matching the record accessor convention — breaking change (see migration guide in README)

Initial release: LTTB, RDP, and PIP downsampling algorithms under a shared Point / DoublePoint data contract
Downsampling static facade as the single entry point
LTTB ported from lttb_downsampling with BucketizationStrategy.DYNAMIC (equal point count) and BucketizationStrategy.FIXED (equal x-span) strategies; LTTB internals namespaced under com.ggalmazor.downsampling.lttb
RDP: iterative (stack-based) implementation; squared-distance comparison avoids sqrt in the hot loop; coordinates extracted into double[] arrays before recursion
PIP: max-heap + array-based doubly-linked list over selected indices replacing the original TreeMap + O(n) scan; prevSel[]/nextSel[] arrays give O(1) neighbour lookup
LTTB DoublePoint fast path uses bucket start/end indices to index directly into pre-extracted coordinate arrays, avoiding getX()/getY() virtual dispatch in the inner loop
JMH benchmarks comparing all three algorithms
Checkstyle (Google Java Style) enforced on main sources