Fix S3 glob XML parsing to ignore non-Contents nodes (e.g. Marker)

[cbardasano]

Summary

This PR fixes a bug in S3::glob where the XML parsing loop incorrectly assumes that all sibling nodes following the first <Contents> node are also <Contents> nodes.

The Issue

When using S3-compatible storage providers (specifically Digital Ocean Spaces in my case), the ListBucketResult XML response may contain additional nodes like <Marker> appearing after the last <Contents> node.

The current implementation uses conNode = conNode->next_sibling() without arguments. This causes the loop to iterate over these non-Contents nodes. Since <Marker> (and probably others) do not contain a <Key> child, the function throws an ArbiterError("Missing Key...").

Example of crashing XML response:

<ListBucketResult ...>
    ...
    <Contents>
        <Key>path/to/file.laz</Key>
        ...
    </Contents>
    <Marker></Marker>
</ListBucketResult>

The Fix

I updated the loop to explicitly request the next sibling with the name "Contents":

// Before
for ( ; conNode; conNode = conNode->next_sibling())

// After
for ( ; conNode; conNode = conNode->next_sibling("Contents"))

This ensures that RapidXML skips any sibling nodes that are not <Contents>, preventing the crash on valid S3 responses that include metadata like markers at the end of the list.