Optimize QR codec and fallback decoding

Add opt-in decodeQR moreEffort retries for rotated, cropped, and upscaled image candidates while preserving diagnostic coordinates and user callback errors.

Speed up codec hot paths with direct bit I/O, cached templates/zigzag paths/RS divisors, faster grayscale stats, monochrome downsampling, and column penalty scans.

benchmark/index.ts vs main: small encodes up to 3.4x faster, large encodes 2.3x faster, generated decodes 1.5-1.7x faster.

Expand decode coverage with PNG vectors, more successful fixtures, and tests for fallback diagnostics, downsampling, and encoder validation.

Author: paulmillr

README.md

@@ -245,25 +245,25 @@ Benchmarks measured with Apple M4 with node.js 25.
 
 ```
 # encode format=ascii
-@paulmillr/qr x 9,050 ops/sec @ 110μs/op
-qrcode-generator x 4,543 ops/sec @ 220μs/op
-nuintun x 3,413 ops/sec @ 292μs/op
+qr x 31,179 ops/sec @ 32μs/op
+qrcode-generator x 4,982 ops/sec @ 200μs/op
+nuintun x 3,735 ops/sec @ 267μs/op
 
 # encode format=gif
-@paulmillr/qr x 8,439 ops/sec @ 118μs/op
-qrcode-generator x 2,909 ops/sec @ 343μs/op
-nuintun x 3,470 ops/sec @ 288μs/op
+qr x 28,111 ops/sec @ 35μs/op
+qrcode-generator x 3,154 ops/sec @ 317μs/op
+nuintun x 3,784 ops/sec @ 264μs/op
 
 # encode of large qr
-@paulmillr/qr x 334 ops/sec @ 2ms/op
-qrcode-generator x 174 ops/sec @ 5ms/op
-nuintun x 221 ops/sec @ 4ms/op
+qr x 1,178 ops/sec @ 848μs/op
+qrcode-generator x 180 ops/sec @ 5ms/op
+nuintun x 237 ops/sec @ 4ms/op
 
 # decode
-@paulmillr/qr x 662 ops/sec @ 1ms/op
-jsqr x 50 ops/sec @ 19ms/op
-nuintun x 49 ops/sec @ 20ms/op
-instascan x 128 ops/sec @ 7ms/op ± 31.44% (4ms..166ms)
+qr x 718 ops/sec @ 1ms/op
+jsqr x 53 ops/sec @ 18ms/op
+nuintun x 51 ops/sec @ 19ms/op
+instascan x 133 ops/sec @ 7ms/op ± 31.77% (3ms..164ms)
 ```
 
 ## License