add blazingly fast instance_cache implementation for K2

[denisichh]

Rewrites instance_cache_store/fetch/update_ttl/delete in the K2 runtime: instead of msgpack-serializing and shipping instances to a separate cache process over RPC, instances are now deep-copied directly into shared memory via the platform's k2::alloc_shared_memory/publish_shared_memory/get_shared_memory.

Storage layout: class_name_hash(u64) | class_instance shell | inner data. On fetch, the block is reinterpreted in place rather than deserialized.

Key pieces

• New "construct-in-place" API on core types (class_instance::clone_in/alloc, array::allocation/copy_in, string::copy_in) -- lets objects be constructed directly into a caller-provided memory span instead of the script memory pool.
• New shared CRTP visitor base kphp::visitors::instance_deep_basic_visitor (runtime-common/stdlib/visitors/instance deep-basic-visitor.h), factored out of the legacy runtime's copy processor so both runtimes share the same traversal logic. Legacy's InstanceReferencesCountingVisitor/InstanceDeepCopyVisitor/InstanceDeepDestroyVisitor now build on it too.
• Two new K2 visitors: instance_deep_copy_visitor (copies an instance graph into a caller-provided block, pinning copies with ExtraRefCnt::for_instance_cache) and instance_deep_estimate_size_visitor (computes the exact size needed upfront so the shared-memory block can be allocated once).
• Alignment: added virtual_builtin_alignof() (dispatches to the dynamic type's alignment for polymorphic clones) plus alignment() helpers on class_instance/array/string, and a padding-aware align_for_chunk(size, align) used with std::align() when carving memory from a pool.
• Removed is_serializable/@kphp-serializable compiler checks for instance-cache calls.