[AdamMagued]

Fixes #170

Problem

On big-endian architectures such as s390x, string escaping produces invalid JSON when strings contain characters that require escaping (such as quotes, backslashes, control characters) within the first four bytes of an 8-byte chunk.

There were two defects in escapeIndex:

1. Word Indexing: In 'for _, n := range chunks', the chunk index was ignored when returning the match index. An escape match in chunk i > 0 returned offset rather than i*8 + offset.
2. Byte Indexing on Big-Endian Architectures: When stringToUint64 casts string memory to uint64 on big-endian systems, byte 0 is positioned in the most significant byte (bits 56..63) and byte 7 in the least significant byte (bits 0..7). Evaluating 'bits.TrailingZeros64(mask&msb)/8' produced 7 - k instead of k, pointing past the character and causing the encoder to emit unescaped characters.

Solution

• When endianness == 1 (big-endian host), reverse the 64-bit word using bits.ReverseBytes64. This normalizes the byte layout so byte 0 is in the least significant byte, matching the LSB-first TrailingZeros64 calculation and arithmetic borrow direction.
• Add chunk index offset i*8 to the returned index.
• Add unit tests in json/escape_test.go validating byte indexing, chunk offsets, edge cases, native execution, and simulated big-endian word layouts.