`sum_value()` / `sum_error()` are not declared `noexcept`

[Bronek]

Neither fn::optional::sum_value() nor fn::expected::sum_value() declares noexcept on any overload.

• The decltype(auto) self-return overloads (include/fn/optional.hpp:495-510, include/fn/expected.hpp:688-703) just return *this by reference and cannot throw — they should be unconditionally noexcept.
• The constructing overloads (optional.hpp:477/486, expected.hpp:670/679) wrap the value in a fresh optional<sum<…>> / expected<sum<…>,E> — they could be conditionally noexcept (keyed on nothrow construction of the sum from the value).
• The free sum_value() (optional.hpp:879, expected.hpp:1020) forward, so they'd propagate the member's noexcept.
• Impact: monadic pipelines lose noexcept unnecessarily.
• Tests: tripwire static_assert(not noexcept(...)) in tests/fn/optional.cpp pin the current behaviour and will flip when fixed.

[Bronek]

Addendum from the #85 test audit: fn::expected::sum_error() has exactly the same gap and should be fixed together with sum_value().

• The decltype(auto) self-return overloads (include/fn/expected.hpp:649-664, and :988-1003 in the expected<void, E> specialization) return *this by reference and cannot throw — unconditionally noexcept.
• The constructing overloads (expected.hpp:631/640, and :970/979 in the void specialization) wrap the error in a fresh expected<T, sum<…>> — conditionally noexcept.
• The free fn::sum_error() (expected.hpp:1024) forwards, propagating the member's noexcept.

Tests: tripwire static_assert(not noexcept(...)) now covers every sum_value and sum_error overload set — tests/fn/optional.cpp (commit 0dc4fcc) and tests/fn/expected.cpp incl. both void-value sum_error shapes (commit 235be7a, branch bronek/unit_tests_cleanup). Each flips green→red when this is fixed.

[Bronek]

Fixed on bronek/bugfixes-for-release-1.0, plus one adjacent defect in the same members.

The noexcept fix, as described. The self-returning overloads are unconditionally noexcept; the lifting overloads weigh both what they construct (the sum) and what they relocate untouched (the other side); the free functions propagate the member's spec.

One consequence worth recording: the lifting overloads still report noexcept(false) even for int, because sum's own value constructor carries no noexcept specifier for them to derive from — the converse half of #280. They sharpen automatically when that lands, and the tests assert both halves so the flip is visible.

Adjacent: they were also the only monadic members not declared constexpr.

static_assert([] {
  fn::optional<int> o{12};
  return o.sum_value().value() == fn::sum{12};  // error: call to non-'constexpr' function
}());

So a constexpr pipeline could not cross the graded lift, while transform / and_then / or_else all can. The free fn::sum_value() / fn::sum_error() lifts were marked constexpr, but forwarded to the non-constexpr members and so were equally unusable. Both are now constexpr, with constexpr twins in tests/fn/optional.cpp and tests/fn/expected.cpp (value, disengaged/error, and free-function paths). Verified on g++ 16.1.1 and clang 22.1.6.

Assisted-by: Claude:claude-opus-4-8[1m]