[AURON #1697] Lazy-cache NativePlan after native conversion check to avoid recomputation

[yew1eb]

Which issue does this PR close?

Closes #1697 .

Rationale for this change

Cache NativePlan nodes after conversion check and reuse in doExecuteNative

What changes are included in this PR?

• Cache serialized NativePlan nodes as lazy vals right after the native support check.
• Reuse the cached nodes in doExecuteNative when creating NativeRDD to avoid recomputation.

Are there any user-facing changes?

No.

How was this patch tested?

[ShreyeshArangath]

Change LGTM, but are we seeing any benefits coming from the lazy-caching? Curious if there is a real-world use-case that drove this change.

[yew1eb]

Change LGTM, but are we seeing any benefits coming from the lazy-caching? Curious if there is a real-world use-case that drove this change.

This change is a minor optimization to avoid unnecessary repeated conversions of Spark Plan nodes (e.g., expressions) into Native Plan nodes (serialized into protobuf format).

[richox]

this native conversions are designed to be eagerly computed in native plan construction, so we can fallback the plan when any conversion fails.

[richox]

this native conversions are designed to be eagerly computed in native plan construction, so we can fallback the plan when any conversion fails.