docs(paper): bring TECHNICAL.tex up to version 2.7.3, and add five figures

[ianktoo]

Two commits. The first brings the technical note's text up to date with
the code; the second gives the note its first figures.

The Zenodo record (v1.1.0, 23 Sep) predates the scrape path and the
v2.6.0 error handling work, so the note no longer described what the
code does. This diffs the paper against HEAD (v2.7.3) and closes the
gaps.

1. Text: up to version 2.7.3

Added: "The pipeline without the LLM" (sec:nollm)

New subsection under System overview. dataforge scrape had only been
mentioned in passing, in the agent guide list and as an MCP table row,
so the capability was effectively undocumented in the paper. It now
covers:

• the same politeness gate as a run (robots.txt, per-domain rate limit,
  Crawl-delay), with no key, session or database
• page title, Markdown and plain text, plus every HTML table extracted
  as a header row plus data rows: the three header forms real pages
  use, colspan and rowspan expanded so the result is rectangular,
  nested tables kept separate, layout tables skipped, spans capped
• the --check rule-based checks (no text, under fifty words,
  duplicate text) and why they are the failures the quality stage would
  otherwise spend LLM calls to notice
• per-table CSV and JSON plus tables.jsonl carrying source_url
• the scrape folder feeding back in as a run's input, so collection and
  generation need not be a single decision

Cost control: the v2.6.0 change

A budget bounds what a working run spends but not what a broken one
wastes. Credential and connection failures are now exempt from retry,
propagate out of the per-chunk handler, and stop dispatch on the first
one. Forty chunks with a bad key now ends in about three seconds with
one message, where it previously made three calls per chunk.

Limitations: two corrections

1. The claim that tabular data is "flattened or lost" stopped being
   true in v2.4.5, but only halfway. The extractor serves the no-LLM
   path, while the dataset pipeline still flattens a table through the
   Markdown conversion. Routing those rows into chunking is now named
   as the reachable half of that gap.
2. The lineage paragraph gained the v2.6.0 recurrence, where a try
   written for one failure class had quietly become the handler for
   every class.

Smaller edits

• Processing says what conversion does to a table and points at both paths
• the MCP section describes scrape_page as a library call beside
  explore_site, rather than leaving it a bare table row
• Discovery cross-references the new subsection
• the abstract gains a sentence on the no-LLM path
• System overview states that the note describes version 2.7.3

2. Figures

The note had four tables and no figures, and three of its arguments are
easier to see than to read. All five are TikZ or pgfplots in the
document itself, so there are no binary assets and they diff in git.

• Figure 1 (System overview): the six stages, what passes between
  them, and where the guarantees attach. Bands mark the three stages
  that overlap as concurrent pools and the two that cannot. Complements
  Table 1 by showing structure rather than repeating its columns.
• Figure 2 (Concurrency): batch vs streaming as a timeline, with
  the LLM idle span marked on the batch row. Drawn schematically and
  the caption says so, because the benchmark ran in streaming mode only
  and the saving is a design rationale, not a measured result.
• Figure 3 (Leak-aware splitting): page to chunks to samples, then
  the same samples under a random split and under a group-aware split,
  with one page marked landing on both sides of the boundary. This is
  the paper's central claim and the one that most wanted a picture.
• Figure 4 (Empirical evaluation): per-site stage timings as
  stacked bars, from
  evals/results/pipeline_metrics_20260922T202407Z.json. Shows USCIS
  bound by its Crawl-delay and the Python tutorial bound by the LLM,
  which the surrounding prose argues.
• Figure 5 (Empirical evaluation): the quality funnel, 831
  generated to 811 approved, with each gate's rejections broken out.
  Carries the ordering claim that the free filters run before the judge
  that costs money.

Every figure is referenced from the prose. Colours were checked for
colour-vision-deficiency separation against a light surface, and every
coloured mark also carries a direct label or a letter, so the figures
survive greyscale printing.

Verification

Compiles clean under tectonic (XeTeX), 24 pages, no errors. Refs,
cites, braces and environments all balance. Every figure was rendered
and inspected, which caught two label collisions the compiler was
perfectly happy with.

Not yet compiled under pdfLaTeX, which is what Overleaf and arXiv run.
To keep that build uneventful: pgfplots compat is pinned to 1.16 rather
than a newer value so it builds on older TeX Live, nothing uses
\includegraphics, and a local style that shadowed TikZ's built-in
thin was renamed. Worth an Overleaf pass before the next Zenodo
upload.

Not in this PR

The DOI in the preamble still points at the v1.1.0 record. A new Zenodo
version mints a new DOI, so that is set at publish time. The concept
DOI in CITATION.cff stays as it is.

The v2.7.1 to v2.7.3 changes are CLI path-resolution fixes and a retag,
judged below the paper's altitude and covered by the version statement
instead.