Reject trailing characters when parsing rfl::Timestamp

[fhgffy]

rfl::Timestamp accepts strings with extra characters after the format:

using TS = rfl::Timestamp<"%Y-%m-%d">;
TS::from_string("1987-04-19T10:00:00"); // ok, time part is dropped
TS::from_string("1987-04-19x");         // ok
rfl::json::read<Person>(R"({"birthday":"1987-04-19x", ...})"); // ok

strptime() returns a pointer to the first character it didn't use, but only a NULL result was treated as an error. This also requires the whole string to be consumed, so these now fail with the existing "did not match format" error. std::chrono::system_clock::time_point parsing already rejects trailing input.

There's also a small fix in the Windows strptime fallback. When get_time reads to the end of the string, tellg() returns -1, so it returned _s - 1. It now returns the end of the string, the same way ParserTimePoint::parse_datetime handles it. I couldn't build on Windows to check that part.

Added json.test_timestamp_trailing_characters. It fails before the change, and the JSON test suite passes after it.

[liuzicheng1987]

@fhgffy this look good to me. Let me run the test pipelines and if they run through cleanly, I will merge.

[fhgffy]

Thanks for running the pipelines. The Linux and Windows workflows passed. The remaining macOS Intel benchmark job stopped while vcpkg was downloading Thrift: curl could not resolve archive.apache.org, before the project was configured or built. The other macOS test and benchmark jobs passed. I cannot rerun upstream jobs with my fork permissions; the failed job may only need a rerun: https://github.com/getml/reflect-cpp/actions/runs/37864987451/job/113823336605

[liuzicheng1987]

@fhgffy OK, I'm re-running