fix: children nodes not carrying metadata from source nodes

[EmanuelCampos]

when running 2 node parsers the metadata is being lost

reproducible example:

from os import path

from llama_index.core.ingestion import IngestionPipeline
from llama_index.core.node_parser import MarkdownNodeParser, SentenceSplitter
from llama_index.core.schema import Document

current_dir = path.dirname(path.realpath(__file__))

sample_pdf_path = path.join(current_dir, "apple_2023.txt")

with open(sample_pdf_path, "r") as f:
    documents = [Document(text=f.read())]

pipeline = IngestionPipeline(
    documents=documents,
    transformations=[
        MarkdownNodeParser(),
        SentenceSplitter(),
    ],
)

nodes = pipeline.run()

with open("output.txt", "w") as f:
    for node in nodes:
        f.write(node.get_content(metadata_mode="all"))
        f.write("\n---\n")

[erik-megarad]

How do we resolve conflicts if there are multiple child nodes all updating the same parent? It seems like it would be good to put it in a separate key and merge when you refer to the nodes. That way you get more control over whether you access only the parent attributes or not.

[erik-megarad]

Oh I think my comment applies to the chunk below. The comment in the code is incorrect- this line is merging parent metadata into the current node's metadata.