[R-312] support adding metadata into testset generation

[jjmachan]

add support for passing metadata into Nodes which exist and returning them as part of the generated testset's reference nodes for further analysis.

ref: https://discord.com/channels/1119637219561451644/1300329780041875517

_R-312

[antonioramos1]

I am adding a document ID to the metadata but since this is lost in the Testset I can no longer know which document was used! Looking forward to have this implemented

[Nimtaa]

As a workaround, I've updated the SingleTurnSample schema to include the sources field (or any other required document metadata).

During sample generation (in the _generate_sample function), while constructing the reference_context, we can also extract the specified metadata from the nodes and include it in the output.

For cases where NodeType.CHUNK nodes are created during splitting and don’t inherit metadata from their parent nodes, we can implement a custom transform that copies the parent metadata to each chunk node. This transform can be applied right after the splitting step.