[w4ffl35]

An untrusted RFC 6902 patch can graft duplicated subtrees via add/copy (or replace the root) to amplify the document past CJSON_NESTING_LIMIT -- e.g. one copy op taking a depth-602 document to depth 1202 -- setting up stack exhaustion in recursive passes.

This refuses grafts whose parent depth plus value depth would reach the limit (new status 12), including root replacement with an over-limit value. The depth helpers are self-bounding: value depth never recurses past the limit, and parent depth is counted from pointer separators (a raw '/' is always a separator since '/' in a key is escaped as ~1).

Evidence:

• Reproducer: copy of a 600-deep subtree under a depth-600 parent now returns 12 with the document unchanged (was status 0, depth 1202); deep root replace refused; shallow copies unaffected. Full ctest suite passes (19/19).
• Adds apply_patches_should_reject_grafts_past_the_nesting_limit and apply_patches_should_allow_grafts_within_the_nesting_limit to tests/misc_utils_tests.c.

Residual note: pure breadth amplification (many small grafts) is memory pressure, not stack exhaustion, and is intentionally out of scope here.