feat: dol.key_in_scope(store, k), a non-writing filt_iter scope probe (#129)

[thorwhalen]

Fixes #129. filt_iter records its predicate on its layer class (_filt_iter_filts); key_in_scope(store, k) follows k through the .store layers outermost first, applies each layer's predicates to the key as it enters that layer (what a write does), maps it with _id_of_key, and finally asks a leaf that has is_valid_key (e.g. Files). No reads or writes. Only a class's own record counts (Store wrappers delegate inner attributes as DelegatedAttribute, which would attribute an inner predicate to the outer layer).

Adversarial review (Opus): agreed with the real write/in for every stacking tried (filt_iter, wrap_kvs, KeyCodecs, Pipe, cached_keys, add_prefix_filtering, cache_vals, mk_dirs_if_missing; class and instance; both orders; set/list/generator filters; subclasses; MI; pickling/deepcopy; Files coordinates). Gaps found and documented in the docstring (no code change): layers that look keys up themselves while keeping the identity _id_of_key (add_path_access, flatten) give a silently wrong answer; the probe raises whatever a layer's _id_of_key raises; a subclass changing the key inside __setitem__.

Dependents gate: pending (the new attribute only touches filt_iter classes).

🤖 Generated with Claude Code