Incorrect byte shift when interpreting 32-bit utf-8 codepoints

[tjanc]

During review of my own JSON string escaping code against popular implementations, I noticed an inaccuracy in yours. Codepoints for 4-byte long utf-8 characters are calculated incorrectly:

    unsigned int calculated = ((firstByte & 0x07) << 24)
      | ((static_cast<unsigned int>(s[1]) & 0x3F) << 12)
      | ((static_cast<unsigned int>(s[2]) & 0x3F) << 6)
      |  (static_cast<unsigned int>(s[3]) & 0x3F);

Notice the byte shift of 24, while there are just 3x6 bits stored in bytes with lower significance.

[cdunn2001]

I wish we had a regression test for this case, but thank you!