Speed up linting of default and examples

[jviotti]

See: #629
Signed-off-by: Juan Cruz Viotti jv@jviotti.com

[augmentcode]

🤖 Augment PR Summary

Summary: Speeds up linting by avoiding work on schema locations/references that cannot be reached during validation.

Changes:

• Bumped vendored dependency SHAs for core, blaze, jsonbinpack, and codegen.
• Added a pointer→location index inside SchemaFrame to make pointer-based traverse() and URI lookup faster.
• Introduced reachability analysis (SchemaFrame::is_reachable) with lazy caching, and used it to skip compiling unreachable reference targets in Blaze.
• Refactored the OrphanDefinitions alter-schema rule to use reachability (and reachable refs-through) instead of scanning ref sets.
• Extended the codegen IR with IRAny so boolean-true schemas compile; TypeScript now emits unknown for that case.
• Updated unordered container usage to rely on explicit Pointer/WeakPointer hashers/comparators instead of a global std::hash specialization.

Technical Notes: SchemaFrame is now non-copyable/non-movable due to internal caches storing references into frame-owned data.

_{🤖 Was this summary useful? React with 👍 or 👎}

[augmentcode]

Review completed. 4 suggestions posted.

Comment augment review to trigger a new review at any time.

[augmentcode]

Making SchemaFrame non-copyable/non-movable is an API-breaking change for any downstream code that stores/returns it by value; please double-check that this is acceptable for the public surface of core.

_{🤖 Was this useful? React with 👍 or 👎, or 🚀 if it prevented an incident/outage.}

[augmentcode]

populate_reachability() memoizes into reachability_ without considering the walker/resolver arguments, so is_reachable() could return stale results if a SchemaFrame is queried with different resolver/walker instances. Is the intended contract that SchemaFrame is only ever used with the same walker/resolver (e.g., the ones used during analyse())?

_{🤖 Was this useful? React with 👍 or 👎, or 🚀 if it prevented an incident/outage.}

[augmentcode]

With the hash implementation now living in GenericPointer::Hasher, dropping the std::hash<GenericPointer<...>> specialization can be a breaking change for downstream users of unordered_*<Pointer> that relied on the default hasher. If this is intended, it may be worth explicitly documenting/calling out in release notes.

_{🤖 Was this useful? React with 👍 or 👎, or 🚀 if it prevented an incident/outage.}

[augmentcode]

This code assumes frame.traverse(reference.first.second) (and the parent traversals) can never return std::nullopt; in NDEBUG builds the assert()s compile out and reference_origin->get() would be UB if the invariant is violated. If unresolved/missing locations are possible here, consider guarding the optional before dereferencing.

_{🤖 Was this useful? React with 👍 or 👎, or 🚀 if it prevented an incident/outage.}