fix: use deterministic hash-based jitter for country centroid fallback

[haosenwang1018]

Problem

The GeoIP centroid fallback uses Math.random() for jitter, causing the same threat from the same country to appear at different coordinates on each request. This creates 'jumping' markers on the visualization.

Fix

Replace Math.random() with a deterministic djb2 hash seeded by the threat ID. Same input always produces the same jitter offset, so markers stay stable across requests.

The jitter range (±1°) is preserved — only the randomness source changes.

Ref: PR #106 re-review (NEW-8)
Fixes #203

[vercel]

@haosenwang1018 is attempting to deploy a commit to the Elie Team on Vercel.

A member of the Team first needs to authorize it.

[koala73]

Thank you @haosenwang1018 can you hanle the below pls :

PR #236 Review: fix: use deterministic hash-based jitter for country centroid fallback

Summary

Two commits: (1) reduce unrest dedup rounding from 0.5° to 0.1°, (2) replace Math.random() jitter in country centroid fallback with deterministic djb2 hash seeded by threat ID. The goal is to prevent threat markers from jumping to different positions on each request.

Issues Found

1. Bug — JSDoc documents wrong output range

The hashJitter comment says "returns a float in [-0.5, 0.5)" but the actual range is [-1.0, 1.0]:

return ((hash & 0x7fffffff) / 0x7fffffff - 0.5) * 2;
//      [0, 1.0]             [-0.5, 0.5]         [-1.0, 1.0]

The * 2 doubles the range. The code is correct (matches the original (Math.random() - 0.5) * 2), but the doc is misleading. Should say [-1.0, 1.0].

2. Redundant commit — unrest dedup change already in main

Commit 1 (83dab23) changes Math.round(lat * 2) / 2 → Math.round(lat * 10) / 10 in unrest/v1/_shared.ts. This exact change was already merged via PR #235 (commit 6f68bb1 on main). This commit is redundant and will cause a merge conflict. The PR needs to be rebased onto current main with this commit dropped.

What's Good

• djb2 is a solid choice for non-cryptographic deterministic hashing — good distribution for marker scatter
• Using index parameter (0 for lat, 1 for lon) ensures independent jitter per axis
• Fallback of seed to countryCode is a sensible default
• Passing threat.id at the call site in hydrateThreatCoordinates gives unique, stable positions per threat

Verdict

Request changes. The deterministic jitter approach (commit 2) is the right solution. Two action items before merge:

1. Fix the JSDoc range comment from [-0.5, 0.5) to [-1.0, 1.0]
2. Rebase onto current main to drop the redundant unrest dedup commit (already merged via fix: reduce dedup coordinate rounding from 0.5° to 0.1° (~10km) #235)