sphinx: avoid repeated isinstance checks

[Dev-iL]

This is a different fix for the same issue addressed in #53364, which was developed in parallel.

The main improvement are:

• Eliminate # type: ignore[attr-defined]
• isinstance branching happens at an earlier scope (so fewer times overall).

Also (according to copilot):

1. Type Safety:

• The suggested approach only processes nodes that are isinstance(node, nodes.Element), which is correct because only Element nodes have .attributes, .children, and .rawsource.
• The existing approach uses node.get(...) and then checks isinstance(node, nodes.Element) later, which is less clear and could allow non-Element nodes to be processed incorrectly.

2. Clarity and Correctness:

• The suggested approach makes it explicit that only Element nodes with the substitution option are processed, matching the docutils node model.
• The existing approach could process nodes that are not Element, leading to possible attribute errors or incorrect behavior.

3. Efficiency:

• The suggested approach avoids unnecessary checks and only runs the substitution logic where it is valid.

[Copilot]

Pull Request Overview

This PR refactors the substitution extension to perform isinstance checks at an earlier scope, reduces redundant branching, and removes a type ignore on the rawsource assignment.

• Move isinstance check for nodes.Element before attribute lookup
• Consolidate per-child isinstance(nodes.Text) logic
• Remove # type: ignore[attr-defined] on node.rawsource

Comments suppressed due to low confidence (1)

devel-common/src/sphinx_exts/substitution_extensions.py:85

• [nitpick] The removal of the type ignore on rawsource assignment changes behavior for some node types. Add tests to verify that rawsource is correctly set for all relevant nodes to prevent regressions in highlighting.

                node.rawsource = node.astext()

[amoghrajesh]

Nice, looks good! Let's wait for CI

[potiuk]

Nice... I've been fighting with it in my Python 3.13 PR as well :D