[numpy] Corrections de style et de terminologie

[Honaminto]

Summary

I went through numpy.md sentence by sentence, comparing it with the English version. In total I made 53 changes in the file, almost all style fixes to make the French read more naturally, plus one small bug fix.

What I changed

Most of my changes fix sentences that stuck too close to the English wording and sounded a bit forced in French. For example, I reworded phrases like "les tableaux disposent de méthodes utiles, toutes soigneusement optimisées" so it sounds more natural.

A few specific things:

• I made terminology more consistent in a few spots ("tableau plat" became "tableau unidimensionnel" in some sentences, to match the vocabulary already used elsewhere)
• I simplified a few clunky sentences in the exercises
• I reworded the intro note on broadcasting so it reads more smoothly
• I found and fixed a small bug: an {index} block was accidentally duplicated under Matrix Multiplication, so I removed the extra one

What I intentionally left alone

A few suggestions I didn't apply, because they either changed the technical meaning of the text or would have created inconsistency with terminology used elsewhere in the file. I also noticed a small mismatch between the text (which mentions tic/toc) and the code (which uses qe.Timer), but that comes from the English version, so I'll flag it as a separate issue instead of fixing it directly in the code.

[mmcky]

Merci Emile — this is exactly the review the edition needed. We validated the PR sentence-by-sentence against the English source (fidelity, French quality, and technical integrity, each independently double-checked): zero regressions across 53 changes, all 109 code cells byte-identical, MyST structure intact. The calque repairs are the standout — « droits égaux », « choquant pour les programmeurs MATLAB », the stranded « de » before the numbered list — and a few edits improve on the English itself (the row/column ordering now matches the code cells that follow). Recommendation is to merge.

Your duplicated {index} find is a defect in the English source (numpy.md lines 432–438 carry the same duplicate) — we're filing it upstream so the next sync doesn't try to fight your fix. Same treatment as your #594.

Four small things before or just after merge, none blocking:

1. Non-breaking spaces: four of the new sentences use a plain space before « : » (« …importantes près : », « doivent être homogènes : », « …unidimensionnel : », « …qu'un seul élément : »). The file's convention is U+00A0 before high punctuation — could you add them (or we'll sweep them in a follow-up)?
2. « tableau plat » survives in 4 places (2 code comments, 2 broadcasting-section sentences). Since we're adopting your « tableau unidimensionnel » ruling into the engine glossary, do you want to finish the sweep here, or shall we do it in the follow-up pass?
3. Two spots to confirm as deliberate: « Le problème essentiel que NumPy résout » became « L'un des principaux objectifs » (the English asserts it as the essential problem), and exercise np_ex1 now forbids only « boucle Python explicite » where the English forbids any form of Python loop (a reader could argue a comprehension is allowed under the French but not the English).
4. « cours » vs « chapitre »: you introduced « chapitre » in a few sentences while « cours » remains elsewhere (and in the other 24 lectures, matching the glossary's Lecture → Cours). Which do you want as the standard term for one lecture? Whatever you pick becomes the glossary ruling for the whole series — happy to do the fleet-wide sweep once you decide.

Also encoded from this PR into the translation engine (so future syncs produce text closer to yours): Tableau unidimensionnel and Masque booléen as glossary pins, plus four style rules generalising your corrections. Thanks for making them minable — the before/after list format is perfect.