Short title (summary): Add C++20 char8_t string accessors to the On-Demand API

Description

PR #2786 added get_u8string() to dom::element. This PR extends the same idea to the On-Demand API, and makes std::u8string_view and std::u8string first-class types in get<T>() on both APIs. Closes #2731.

Every string simdjson produces is valid UTF-8, so all of the new accessors are zero-copy aliases of their char-based counterparts: the same bytes, at the same address, no conversion.

New accessors

Type-based access

• get<std::u8string_view>() on ondemand::value, ondemand::document, ondemand::document_reference and dom::element — the DOM specialization also brings dom::element::get(std::u8string_view&) and is<std::u8string_view>() along for free.
• get<std::u8string>(), plus containers and optionals of it (std::vector<std::u8string>, std::optional<std::u8string>, …), through the C++20 deserialization path. This is a new concepts::constructible_from_u8string_view concept and a matching tag_invoke overload; std::u8string cannot use the existing constructible_from_string_view path because the character types differ.
• The templated receiver overloads get_string(receiver) and unescaped_key(receiver) now accept a std::u8string or std::u8string_view receiver. Previously they did not compile for those types, because the implementation did a plain receiver = content; from a char-based std::string_view. That single assignment is now routed through an internal::assign_utf8() helper whose general (unconstrained) overload is the old code verbatim, so no existing receiver type changes behavior.

Notes

• A new SIMDJSON_SUPPORTS_CHAR8_T macro in compiler_check.h replaces the raw #if defined(__cpp_char8_t) && __cpp_char8_t >= 201811L feature test that PR Add C++20 u8 variants of string access methods (#2731) #2786 repeated at every call site. It follows the style of the neighbouring SIMDJSON_SUPPORTS_CONCEPTS / SIMDJSON_SUPPORTS_DESERIALIZATION macros, and is user-overridable.
• Everything is purely additive and guarded, so C++11/14/17 builds are byte-for-byte unaffected.
• get_wobbly_string() deliberately has no u8 variant. Wobbly strings may be invalid UTF-8 (WTF-8), so they have no business in a std::u8string_view.
• No serialization/builder changes: this issue is about string access.

Type of change

• Bug fix
• Optimization
• New feature
• Refactor / cleanup
• Documentation / tests
• Other

How to verify / test

• New test target tests/ondemand/ondemand_u8string_tests.cpp (17 tests): value/document/document_reference/field accessors, the allow_replacement flag, INCORRECT_TYPE and error forwarding, iterate_many, get<T> for both u8 types, containers and optionals, u8 receivers, a byte-for-byte check that the u8 view matches get_string(), and a regression check that the char-based paths still behave.
• tests/dom/basictests.cpp gains u8string_value_get_template() covering dom::element::get<std::u8string_view>(), get(T&), is<T>() and the error path.
• Full suite passes under both standards on Apple clang 17 (arm64): 121/121 with SIMDJSON_CXX_STANDARD=20 and 121/121 with SIMDJSON_CXX_STANDARD=17 (where the whole feature compiles out).
• Also compiled the library and a driver under -std=c++11, -std=c++14, -std=c++17 and with -DSIMDJSON_EXCEPTIONS=0.

Checklist before submitting

• I added/updated tests covering my change
• Code builds locally and passes my check
• Documentation / README updated if needed (new u8string_view and u8string (C++20) section in doc/basics.md, note in doc/dom.md)
• Commits are atomic and messages are clear
• I linked the related issue

https://claude.ai/code/session_01EKoa1vNn8ZKwJH4qaZTU2M