gettext: Remove obsolete source metadata workarounds

[jdillard]

Purpose

Docutils now provides source and line metadata for the following nodes across all versions currently supported by Sphinx (>=0.21,<0.23):

• Literal blocks nested under bullet lists
• Rubric directives
• Images in substitutions and nested directives

This removes the corresponding obsolete source-metadata workarounds and the regression test that only exercised the removed literal-block fallback. Existing gettext and message-extraction tests continue to cover these cases.

The workarounds for line nodes and field names remain because supported Docutils versions still omit their metadata.

References

• Add .. rubric:: to the list of special cased nodes for gettext. #1305
• gettext build broken on substituted strings #3093
• translation fails when literal block is in a list  #4913
• Fix #4913: i18n: literal blocks in bullet list are not translated #4943
• 6.1 breaks sphinx-argparse-cli #11091
• Docutils feature request #41: Not all Node instances have the source and line attributes set

AI Disclosure

AI was used to investigate whether the workarounds remain necessary and draft the implementation. I reviewed and validated the changes.