GFQL polars general path drops null-id node rows and collapses duplicate ids on node-only op lists (pandas keeps them)

[lmeyerov]

Found by

The route harness (frame variants, PR #2070) on the productized polars single-node route.

Repro

Node frame with a null id row (nullable Int64) or a duplicated id row; op list [n()] or [n({"w": GT(2)})].

frame	engine	lane	general path
null-ids	pandas	[1,2,3,4,5,None]	[1,2,3,4,5,None]
null-ids	polars	[1,2,3,4,5,None]	[1,2,3,4,5]
dup-ids	pandas [n()]	[1,1,2,3,4,5]	[1,1,2,3,4,5]
dup-ids	polars [n()]	[1,1,2,3,4,5]	[1,2,3,4,5]

The pandas oracle keeps both; the polars single-node lane matches it; the polars general chain drops the null-id row and collapses the duplicate. Related: #2034 (duplicate-id contract on node lookups vs hops).

Where

graphistry/compute/gfql/lazy/engine/polars/chain.py general forward/backward passes for a node-only op list (the endpoint/closure joins never link a null id and dedupe by id, which is right for hops but not for a bare node lookup).

Ledger

("polars-single-node", "null-ids") in the route harness KNOWN map until fixed.