[P1] Use semantic key identity in borrowed Node merges

[jskoiz]

Summary

• Make borrowed &Node merge expansion use same_key_identity, matching the owned merge path and ignoring source spans/source metadata.
• Propagate semantic-key depth-limit errors from borrowed merge insertion.
• Add regressions for distinct-span scalar keys, integer/unsigned-equivalent keys, and reordered equivalent mapping keys across from_node, borrowed &Node, and owned Node deserialization. The tests preserve explicit/local-key precedence.

Verification

Local proof:

• git diff --check — passed.
• cargo fmt --all --check — passed.
• Focused test attempted: cargo test --locked --test serde_value_api serde_api_borrowed_node_merge_uses_semantic_key_identity.
• The focused Cargo test did not complete because the shared filesystem ran out of space during dependency compilation: No space left on device (os error 28). Per the task handoff, no further Cargo builds were started.

Hosted/merge state:

• Branch pushed to origin as c1/issue-76-borrowed-merge-identity.
• Current observed PR state: draft/open against main; CI run 32536800326 has six Rust/audit jobs in progress, and GitGuardian is successful.
• Hosted CI completion is not claimed.
• This PR has not been merged.

Notes/Risks

• Full tests, clippy, and other build-backed gates remain pending because the host had approximately 125 MiB free after the failed compile.
• The change is limited to borrowed Node merge-key identity and its focused regression coverage; the owned merge path and existing YAML merge precedence are otherwise unchanged.