Fix duplicate MERGE default projection

[adsharma]

Fixes: #170

Summary

This fix separates the responsibilities more cleanly:

Planner / LogicalMerge:

• Decides whether duplicate-created MERGE rows are user-visible.
• Computes suppressDuplicateCreatedOutput from the logical child schema, merge keys, inserted pattern expressions, and ON CREATE / ON MATCH state.
• Suppresses duplicate-created output only when the row has no non-key upstream payload and no ON CREATE side effects.

Mapper / map_merge:

• No longer classifies payload semantics.
• Just reads logicalMerge.suppressesDuplicateCreatedOutput() and passes that decision into physical MergeInfo.
• Still owns the mechanical translation from logical expressions to evaluators, executors, DataPos, and local table layout.

Executor / persistent Merge:

• Enforces the planner’s decision.
• The pattern creation table still deduplicates physical creation by merge key.
• If a later row hits an already-created key, the executor either suppresses the duplicate-created output row or rehydrates the already-created pattern for output, depending on the planner-provided flag.

The important semantic invariant is:

MERGE creates at most one physical pattern per merge key, but duplicate-created output rows are preserved only when they carry user-visible upstream payload.

That gives the desired behavior for both cases:

• UNWIND [1, 1] AS i MERGE (a {stuff: i}) RETURN a.id returns one row.
• MATCH ... WITH ..., age MERGE (...) RETURN u.ID, age still returns one row per upstream payload row.

The UUID/default part is fixed separately but consistently: when a duplicate-created row must still be output, we read the already-created node’s projected properties from storage instead of re-evaluating default expressions. That prevents volatile defaults like gen_random_uuid() from producing a fresh value for a row that was not actually inserted.

Root Cause

The unwind dedup changes let duplicate no-match MERGE rows continue through the merge output path. The duplicate-created path skipped the physical insert but still evaluated insert defaults again, so projected defaults such as gen_random_uuid() differed from the row that was actually stored.

Validation

• cmake --build build/release --target lbug_shell e2e_test --config Release
• manual release-shell repro for duplicate UUID MERGE
• E2E_TEST_FILES_DIRECTORY=test/test_files build/release/test/runner/e2e_test --gtest_filter='issue~issue.issue_170'
• E2E_TEST_FILES_DIRECTORY=test/test_files build/release/test/runner/e2e_test --gtest_filter='issueissue.unwind_merge_rel_after_rel_match_dedup:issueissue.unwind_merge_with_projection_dedup:issue~issue.unwind_merge_with_projection_preserves_match_expansion'