fix: resolve 32-bit truncation overflow in BigInt coordinate math

- Replaced 32-bit bitwise shifts (<<) with Math.pow() and BigInt shifts to support large nsides without modulo overflow.
- Recalculated compress_bits and spread_bits to explicitly loop 64-bit indexes rather than truncating with 8-bit Uint16Array table lookups (ctab/utab).
- Restored original C/C++ geographic precision boundaries allowing the engine to successfully encode and decode high-precision pixels natively up to Order 52.

package.json

@@ -1,6 +1,6 @@
 {
     "name": "@gkucmierz/healpixjs-bigint",
-    "version": "1.0.1",
+    "version": "1.0.2",
     "description": "True BigInt port of healpixjs supporting orders beyond the 32-bit (<=14) limit",
     "keywords": [
         "healpix",
@@ -31,4 +31,4 @@
     },
     "author": "Grzegorz Kućmierz",
     "license": "MIT"
-}
+}