Encode uint64 values above INT64_MAX in UBJSON

[youdie006]

encode_ubjson writes nothing for a uint64_t value above INT64_MAX but still counts it as a container element. [9223372036854775808] becomes 5b235501 (an array that declares one element and contains none), and a top-level value becomes an empty buffer; decode_ubjson and other UBJSON readers reject both.

UBJSON has no unsigned 64-bit type, so this writes such values as high-precision numbers (H), as the encoder already does for bigint strings; the parser reads them back as semantic_tag::bigint. The bytes match py-ubjson's encoding of the same values.

Added a test with the exact bytes and a decode_ubjson round trip, plus a guard that INT64_MAX stays L. The ubjson tests pass under ASan/UBSan.

Written with AI assistance (Claude); I have reviewed the change.

[danielaparker]

Thanks for contributing!

I would suggest replacing

for (auto c : s)
{
    sink_.push_back(c);
}

with

sink_.append(s.data(), s.size());

Otherwise it's fine.

[youdie006]

Thanks, done in d5f5fcf. The bytes sink's append takes const uint8_t*, so it casts the way cbor_encoder does: sink_.append(reinterpret_cast<const uint8_t*>(s.data()), s.size()). The ubjson tests still pass.