maintenance: raw-string the LaTeX-carrying docstrings

[Jammy2211]

Part of PyAutoLabs/autolens_workspace#491 — one of six independent, prose-only PRs (one per workspace repo). No API surface, so no cross-repo merge ordering.

This repo carries the defect the whole task was filed on. In chapter_4_scaling_up_lensing/tutorial_3_scaling_relation.py, \theta_E was stored as TAB + "heta_E" — a silent corruption with no diagnostic whatsoever. Confirmed repaired: the docstring value now holds a real backslash and contains no TAB.

	what it is	diagnostic
warned	\s, \l, \[ — escapes Python does not recognise	SyntaxWarning per compile; slated to become SyntaxError
silent	\t in \theta, \f in \frac, \r in \rm	none at all — the value is simply corrupted

chapter_4_scaling_up_lensing/tutorial_5_cluster_scale.py has only silent hits and zero warnings, so a warning-only sweep skips it entirely. It is included here.

32 literals across 8 files get the r prefix. Both sweeps now return zero. Prose is untouched — only the delimiter gains an r.

Verification

• Baseline regeneration is a no-op — the generator was run before editing and left the tree clean, so generator noise cannot fake the gate below.
• Runtime values: 21 corruptions repaired, 0 other changes. Every changed literal's value was compared HEAD vs worktree; the prefix may only ever remove corruption. This is the check that catches real mistakes — the regeneration gate structurally cannot, because the generator reads source text, not runtime values.
• Regenerated with autohands: notebooks/, markdown/, llms-full.txt and workspace_index.json are all byte-identical — the diff-empty gate passes exactly.

Two docstrings needed a hand-approved prefix, both read in full first: chapter_2_lens_modeling/tutorial_1_non_linear_search.py has pre-existing malformed nested LaTeX the math-span matcher cannot bracket, and tutorial_3_scaling_relation.py writes bare \sigma/\theta_E/\pi in undelimited equation blocks. Neither carries a deliberate escape. Malformed delimiters were left exactly as found.

Left alone deliberately

Real newlines in print() and already-escaped LaTeX line breaks. The prefix was applied only where every backslash in the literal sits in a LaTeX context ($…$, \(…\), \[…\], \begin{}…\end{}, or a markdown code span), with an absolute veto on \\, escaped quotes and numeric escapes regardless of context.

Unblocked by PyAutoHands#251, which taught the notebook and env parsers to accept an r""" opener.

🤖 Generated with Claude Code

https://claude.ai/code/session_01MTjtx5mdituitiyQYFGn2E

---

Generated by Claude Code

[Jammy2211]

Smoke is red on this PR, but the failure is on main, not here — merging despite the red check on that basis.

Two scripts fail with numba Pass nopython_type_inference:

chapter_3_pixelizations/tutorial_8_adaptive_pixelization.py
chapter_3_pixelizations/tutorial_11_brightness_adaption.py

Evidence it isn't this PR:

• Neither file is in the diff, and all 8 files that are pass smoke — including tutorial_5_bayesian_formalism.py in the same directory.
• The failure reproduced identically on one re-run.
• Control experiment: I re-ran the last green main build (run 255, commit dcb67e9, zero changes from this branch, fresh dependency install). It now fails on the same script with the same error — the Slack payload from that run reads HowToLens/main smoke tests (Python 3.12) failure. So main itself is red.
• A one-character r prefix on a docstring cannot cause a numba typing failure.

Likely cause, not proven: PyAutoLabs/PyAutoArray#453 ("in-place Cholesky buffer + copy-free numba solves for fnnls_cholesky") merged at 22:09:30 UTC, minutes before these runs. CI builds autoarray from source at HEAD, and the last green main run (04:29 UTC) predates it. fnnls is the NNLS solver pixelization reconstruction uses, and the two failing scripts are the adaptive-pixelization ones.

I tested two specific mechanisms under the exact CI versions (numba 0.67.0 + scipy 1.17.1) and both passed — a strided Ubuf view into _cholupdate, and the newly-introduced np.dot inside _cho_solve_buffer (correct to 2.8e-17). I also ruled out a new numba release: 0.67.0 shipped nine days before the last green run. So the attribution rests on timing and code area, not a reproduced mechanism — worth a look from someone who owns that solver.

This PR's own verification is unaffected: 21 corruptions repaired with 0 unexpected value changes, and notebooks/, markdown/, llms-full.txt and workspace_index.json all byte-identical after regeneration.

---

Generated by Claude Code