Docs: clarify VLE semantics and cost model (#2349)

[gregfelice]

Summary

Adds an in-source design-note comment block at the top of src/backend/utils/adt/age_vle.c documenting:

1. Semantics — edge-isomorphism per openCypher: no repeated edges per path, vertices may recur. Includes a small triangle example showing why a visited-node ("vertex-isomorphism") optimization would silently drop spec-valid paths.
2. Cost model — bounded [*min..max] is polynomial in edge count for fixed max; unbounded [*] / [*1..] is factorial worst-case on cycle-rich graphs. This is inherent to edge-isomorphic enumeration and cannot be reduced without changing semantics.
3. Implementation pointer — one paragraph identifying edge_state_entry.used_in_path, dfs_find_a_path_between/dfs_find_a_path_from, and is_edge_in_path() as the cycle-prevention enforcement sites.

Motivation

Issue #2349 misdiagnosed the VLE as lacking cycle detection and proposed a visited-node filter as a fix. Cycle prevention already exists via edge-uniqueness, and the proposed change would violate the openCypher specification. The current source has no overview comment explaining these semantics, so a future reader would have to reconstruct them from the DFS code and the spec. This comment block captures the semantics and cost model at a point the reader is guaranteed to pass through, preventing repeat misanalysis.

Per @jrgemignani's guidance on #2349, no GUC, no code change — just documentation.

Scope

• Files changed: 1 (src/backend/utils/adt/age_vle.c — 44-line comment block inserted between the license header and the first #include).
• No behavior change, no new symbols, no tests added (nothing to test).

Test plan

• Cassert build (--enable-cassert PG18) clean, no new warnings
• Cassert installcheck green — 33/33 AGE tests pass, including cypher_vle