Add UUID conversion to and from 16 byte fixed sequences

[urmastalimaa]

• Created upstream PR [Proposal]: Add UUID conversion to and from 16 byte fixed sequences beam-community/avro_ex#100
• If there's no interest or activity, I'll merge into our fork to unlock optimizations for our applications

---

UUIDs are often passed around in application code in their canonical,
hex as string representation e.g. "550e8400-e29b-41d4-a716-446655440000".
Encoding UUIDs as Avro "string"s takes 37 bytes, while encoding UUIDs in
their binary form fits into a 16 byte sized "fixed", saving 21 bytes per
encoding.

This change allows application code to keep passing around canonical hex
UUIDs while converting to the compact encoding, requiring only
uuid_format: :canonical_string to be given in decode options.

The Java reference implementation also supports
encoding UUIDs as both strings and 16 byte fixed sequences.

• Encoding is augmented such that a 16 byte fixed schema with
  %{"logicalType" => "uuid"}, converts a hex-string UUID to the 16
  byte binary representation.

• Decoding is augmented such that given uuid_format: :canonical_string
  in decode options, the binary representation is converted to the
  canonical hex-string representation.

The encoding change is nearly backwards-compatible, previously when
given an incorrectly size "fixed" with {"logicalType": "uuid"}, an
error was raised, while now conversion is attempted.

The decoding change is fully backwards-compatible, as uuid_format
defaults to :binary.

For UUID codec, the uniq library was added (no transitive
dependencies).

[indrekj]

This makes sense to me. cc @laurglia you were asking about the same thing some time ago