[ngallodev]

Summary

• Decode persisted LSP definition and string arrays with yyjson_arr_iter to avoid repeated scans of non-flat arrays.
• Add a regression check for definition order, null values, and empty parameter arrays.
• Adapt upstream codebase-memory-mcp commit d7eba5a7 for the CLI fork.

Review and validation

• Source review found no actionable correctness issues.
• CBM_NO_CCACHE=1 scripts/build-dev.sh passed.
• ASAN_OPTIONS=detect_leaks=0 CBM_NO_CCACHE=1 scripts/test.sh --suites pipeline passed: 274 tests. This validation ran before the request to stop running tests; no tests were run after that instruction.
• git diff origin/main...HEAD --check passed.
• CodeRabbit review could not start because browser authentication is unavailable in this environment.

This PR does not claim Agent-Workflow evaluation, acceptance, or merge approval.

[ngallodev]

Reviewed, no issues found.

Checked by reading the PR head of src/pipeline/lsp_surface.c:

• yyjson_arr_iter yields the same elements in the same order as yyjson_arr_get(i) for i in [0, size). Duplicates and nulls are preserved, and the iterator count matches yyjson_arr_size, so the i index never passes the allocated count. It cannot overrun the items/defs allocations.
• Empty and missing arrays behave as before: an empty lsp array returns 0 early, and an empty spt array gives count 0.
• Non-string items still map to "?". Null-terminator placement (items[count]) is unchanged.
• Error paths are unchanged. The arena owns the allocations, and the doc is freed on every early return. Allocation-failure handling (NULL from cbm_arena_strdup) is the same as the old code.
• The test covers order, null receiver, null param item and the empty array. It checks behavior only and cannot detect a regression back to O(n^2). Fine for a perf-only fix; a large-input timing check would be the only way to cover that.

Tests and builds not run by me.