fix(labels): handle missing metadata and fix add_labels serialization

[geopanther]

Summary

Fix two label-related edge cases that cause crashes on some Confluence server versions.

Adapted from iamjackg/md2cf#118 by @nate-woythaler.

Changes

1. add_labels(): Changed from data= (form-encoded) to json= for proper JSON serialization of label payloads
2. labels_need_updating(): Added defensive getattr chain to handle pages without metadata.labels.results — treats missing metadata as 'needs update'

Differences from upstream

• Upstream changed labels_need_updating() to use dict-style .get() access, which breaks the existing Bunch-based object model. We use getattr chains instead, preserving compatibility with the Bunch response objects
• Upstream's json.dumps() wrapper for update_page labels is unnecessary since requests handles JSON serialization via the json= parameter — omitted
• Removed dead # return self.api.content(page.id).post( comment

Tests

• New test: test_labels_need_updating_when_no_metadata
• All 17 upsert tests pass

[geopanther]

Maintainer Review

Verdict: ✅ Approve

Both fixes address real bugs that would hit users of self-hosted Confluence or pages created before labels were supported.

add_labels fix: Using data= with a list of dicts sends form-encoded data, which Confluence can't parse as JSON. Switching to json= is correct and matches how update_page already sends label data.

labels_need_updating fix: The triple-getattr chain is defensive but appropriate — we can't control what the Confluence API returns, and missing metadata is a documented issue on older versions. Treating missing metadata as 'needs update' is the safe default.

vs upstream: Our approach is cleaner — upstream switched to dict-style access (.get()) which is inconsistent with the Bunch-based object model used everywhere else. We preserve the object-attribute access pattern.

No concerns. Ship it.