Fix UB from error out-params not written on success paths

[harmony7]

The bridge error convention declares the out-param uninitialized on the C++ side and relies on the Rust side to write it. try_fe! (and the equivalent try_kv! / try_is! macros) do set it to null on success, so the pattern is sound for any function that routes every path through a macro.

Two KV store functions bypass those macros and wrote err only on their error path, leaving the success path reading an indeterminate pointer:

• ListResponse::next() (src/kv_store.rs) - the Ok(page) arm set out only.
• m_static_kv_store_kv_store_open() - the Ok(..) arms set out only.

When the stack slot happened to be non-zero, the C++ wrapper saw err != nullptr, reported a bogus error, and called rust::Box::from_raw() on garbage, which would later drop a garbage box. Cross-language LTO happened to leave those slots zeroed, so the test suite passed; building with LTO off surfaced it as a failure of REQUIRE(page.has_value()) at test/kv_store.cpp:92. Both outcomes were luck rather than correctness.

Both functions now null err on entry, which also covers the paths that return false without an error at all: iteration finishing in next(), and Ok(None) (store does not exist) in open(), which the C++ side must be able to tell apart from a real failure.

Also gives all 67 error out-param declarations across include/ and src/cpp/ an explicit {nullptr} initializer. That is redundant where a try_*! macro already guarantees the write, but it makes the convention robust by default so a future bridge function that forgets cannot reintroduce this class of bug.

Verified by building with LTO disabled, where kv_store previously failed and now passes (7/7); the normal LTO build and lint gates are unaffected.