A NotebookLM video overview, so the workshop sells itself before anyone arrives

[project-delphi]

Blocked by #9, #39, #43, #44 and #45 — do not start until all five are
closed. The video is a recording of the workshop as it finally is. Making it
while the notebooks (#44), the decks (#45) and the figures (#9) are still
moving means making it twice, and a stale video is worse than no video because
people trust it.

Why

Participants arrive having been asked to do 2–3 hours of prerequisite work
(linear-algebra-deep-learning) before a 195-minute session. Right now the only
thing that sells that ask is a wall of text on the landing page. Nothing on the
site says this is worth your Saturday in a form someone will actually consume
on a phone the night before.

A short video overview fixes that, and it is now cheap to make: NotebookLM will
generate one from sources rather than from a prompt, so what it says is grounded
in this repo instead of invented. That property is the whole point — the same
reason every figure here is drawn from an array the workshop actually uses.

The ask

@Laverde97 — build a NotebookLM notebook over the sources below, generate a
Video Overview, upload it, and link it from the handbook and the site.

1. The sources to load

Everything in this repo that is prose or teaching content:

• tensors_workshop_plan_with_quizzes.md — the handbook, including
  Appendices A–E. This is the spine; load it first.
• The rendered site pages: index.qmd, notebooks.qmd, kahoot.qmd. Feed
  the published URLs (https://project-delphi.github.io/tensors-workshop),
  not the .qmd source — the {{< var >}} shortcodes are unresolved in
  source and NotebookLM will read them literally.
• README.md and notebooks/README.md.
• The twelve notebooks. If NotebookLM will not ingest .ipynb directly,
  export them (docs/notebooks/*.ipynb are the served copies) or point it
  at the Colab links.
• The three Kahoot spreadsheets, so the quiz topics are represented.

External references for tensors, so the video can place the workshop in a
literature rather than talk only about itself:

• Deep Learning (Goodfellow, Bengio & Courville), Chapter 2 — the assumed
  prerequisite, already workshop.book in _variables.yml.
• Laverde97/linear-algebra-deep-learning — the required pre-work.
• Kolda & Bader, Tensor Decompositions and Applications (SIAM Review 2009)
  — the standard reference behind sections 10 and Appendix C.
• The TensorLy docs, and NumPy's einsum documentation.

Add anything else you think earns its place, but keep the ratio honest: this
repo's own content should dominate, or the video will drift into generic
"what is a tensor" material that the workshop deliberately does not spend time on.

2. What the video needs to do

It is a trailer, not a lecture. Target 3–5 minutes.

• Motivating, not descriptive. Lead with what someone can do at the end
  that they cannot do now — manipulate, solve, convolve and factorize
  tensors — not with a table of contents.
• Say the data is real. Real tumour measurements, real handwritten
  digits, real histology, real NYC taxi trips, real airline traffic. That is
  the single most distinctive thing about this workshop and the reason the
  exercises have problems in them worth finding.
• Set the honest expectation: no prior tensor theory needed, but the
  linear algebra pre-work is not optional and the session moves fast.
• Name the shape of the session — twelve sections, three Kahoot checks,
  everything in Colab with nothing to install.
• Check it for errors before publishing. Generated narration will invent
  confident detail; anything it says about shapes, datasets or the running
  time has to match _variables.yml and the notebooks.

3. Where it lands

Host it unlisted-but-linkable (YouTube unlisted is fine) so it survives outside
NotebookLM, and so the link is stable.

• Add a video: block to _variables.yml — the URL, title_en/title_es,
  and its length. Do not hard-code the URL anywhere else. Follow the
  kahoot: precedent: land the key with a placeholder and a # TODO comment
  if the upload lags the code change.
• Surface it near the top of the handbook, above The Data We Use — the
  handbook is in the render: list, so {{< var >}} resolves there.
• Surface it on index.qmd in What this is, and on es/index.qmd in
  lockstep. Bilingual, like every other string.
• Slides: optional, and only if it does not cost time on the clock. If it
  goes in, it belongs on the §00 title slide as a link for people to watch
  later, never as something played during the session.
• quarto render and commit docs/ in the same change — Pages serves the
  committed docs/, and compare_render.py will go red otherwise.