Document on btck_chainstate_manager_options_set_wipe_dbs that a wipe must be followed by btck_chainstate_manager_import_blocks before the chainstate manager is used for anything else, as the existing kernel tests already do (specifically the chainman_reindex* kernel tests). Note in the @return of btck_chainstate_manager_get_best_entry that it can return null when no block headers have been loaded.
Background: I've been working on a bindings project using the libbitcoinkernel and tripped on a SIGSEGV calling the entry accessors on the null pointer returned by get_best_entry after a (true, true) wipe (#35293). The C++ wrapper handles this via btck::check<>, but bindings generated from the C header don't see the wrapper. Adding a nullability documentation hint helps generators produce the correct signature, and the @note on set_wipe_dbs documents the lifecycle that avoids the null in the first place.
Docs-only change; no tests ran.
