Prevent include parsing across inline documentation tokens

[RyanTanuki]

Summary

Fixes #1974 by preventing the body-form include regex from crossing a nested <include opener.

Root cause

An inline documentation token could pair with a much later genuine closing tag. The parser returned hundreds of lines as one include path, and agentic_update._compute_include_allowlist raised OSError: [Errno 36] File name too long when constructing Path(...).

Changes

• Temper the body-form regex so a match cannot cross another include opener.
• Preserve bare, attributed-body, self-closing, and include-many parsing.
• Add the exact inline-token-before-real-include regression.
• Update the owning prompt contract and run deterministic PDD metadata sync.

Validation

Per the requested workflow, no local test command was run; PR CI is the validation gate.

[RyanTanuki]

Added commit 115dbaedc to record the PDD-generated sync_order fingerprint.

Root cause of the first auto-heal failure:

• The parser prompt and generated code were intentionally changed together.
• Deterministic metadata sync generated .pdd/meta/sync_order_python.json, but the repository-wide metadata ignore rule left this previously untracked module fingerprint out of the PR.
• Without a tracked current fingerprint, auto-heal correctly reported an unresolved prompt/code conflict.

The fingerprint is now explicitly tracked as the reconciliation record. The other reported drift, agentic_change_orchestrator, pre-exists this branch and is repaired by PR #1620; this branch will merge updated main after #1620 lands.