[Flow Validate] Add canonical-output linting for finalized vs intermediate results

[ameet]

Summary

The flow engine would benefit from a notion of a canonical final output step, plus validation/linting around pre-final result consumption.

Problem

In non-trivial flows there is often a progression like:

• buildResult
• optional enrich/merge steps
• persistence or upload steps
• finalizeResult

Today, parent flows and cache steps can accidentally read an intermediate output instead of the finalized one. This is especially easy when the flow technically runs, but the consumer misses persisted URLs, normalized fields, or merged data that only exist after finalization.

Requested behavior

A few options could help:

• support a canonical output step concept, inferred from finalizeResult or declared explicitly in flow metadata
• warn when buildCacheEntry caches a pre-final result even though a canonical final output exists
• warn when a parent flow reads a child flow's intermediate step output instead of the canonical final output

Why this would help

This would catch bugs that are otherwise very subtle:

• cached results missing persisted fields
• parent flows consuming incomplete child data
• output shape drift between fresh runs and cached runs

Possible design direction

Even a lightweight lint would be valuable. For example:

• default canonical step order: finalizeResult > buildResult
• optional flow metadata override such as canonicalOutputStep
• validator warnings first, hard errors later if the heuristic proves reliable

This seems like a good fit for flow validate because the engine already knows step ids and execution order.