Use simdutf for faster string encoding/decoding

[wh201906]

This PR migrates some string encoding/decoding functions in HybridUtils from OpenSSL/manual implementations to simdutf:

• Add the full source of simdutf v8.2.0
  • The source code size could be optimized by generating it with limited features, but I didn't find significant size increases from the release build apk of example after introducing the whole library.
• Use simdutf for base64/base64url encoding and decoding
  • The OpenSSL implementation doesn't support SIMD instructions on mobile devices yet. Seems like it only supports AVX/AVX2 encoding/decoding. (Issue 29739/optimize avx base64 decoding openssl/openssl#30610, ISSUE-29739: Optimize base64 decoding when AVX2 is available openssl/openssl#30499, Improved Avx2 + Scalar Base64 Encoding openssl/openssl#29178)
• Use simdutf for encodeLatin1()
  • The SIMD stuff should provide higher speed, although the benchmark doesn't cover it yet.
• Remove unnecessary intermediate data copies in decoding flows
• Add some decoding test cases for the new behavior, which is aligned with Buffer.from() in Node.js

These changes significantly improve performance when encoding/decoding large base64 payloads. I ran the benchmark on an old Android device and here are the results:
(I added some test cases with 1kB data, but this PR doesn't include them)
base64 1MB encode throughput compared with CraftzdogBuffer: From 4.92x to 9.16x
base64 1MB decode throughput compared with CraftzdogBuffer: From 169.07x to 398.76x

• old

• new

[wh201906]

There are some ongoing optimizations in simdutf, which can also be utilized in RNQC once implemented.
simdutf/simdutf#925
simdutf/simdutf#565