Optimize planner stats and functional recursive joins

[adsharma]

Summary

This PR adds generic planner stats support for relationship cardinality estimates, an explicit ANALYZE command to refresh planner stats, and a guarded recursive-join fast path for functional chains.

The functional-chain recursive join fast path applies to a narrow, generic case:

• variable-length walk (*min..max)
• forward direction only
• one node table on both ends
• one relationship table from that table back to itself
• no path materialization
• each source node has at most one valid outgoing edge

When those conditions hold, the relationship graph is a set of chains or cycles where every node has zero or one next node. Instead of running the full recursive graph-search machinery, the operator first builds a nextOffsets array:

node offset -> next node offset, or INVALID_OFFSET

If it sees two outgoing neighbors for any node, it bails out and uses the normal recursive implementation.

Once the array is built, each source node is handled by pointer chasing:

src -> next -> next -> next ...

It emits (src, dst, length) rows for lengths in the requested recursive range, respecting lowerBound, upperBound, input/output masks, visibility, and limit counters.

Why this is faster: the normal recursive machinery has to manage frontier expansion, graph scans, path semantics, intermediate state, and generalized output handling. For a functional chain, there is no branching, so the next hop is just an array lookup per step. That turns recursive expansion into tight linear pointer chasing while preserving the same result shape for the guarded case.

Validation

• cmake --build build/release --target lbug_shell -j 8
• RE24 script against /tmp/bb_full_hash_off_fwd_art.lbug in read-only mode
• ANALYZE; and ANALYZE Event; against /tmp/bb_full_hash_off_fwd_art.lbug in read-only mode