Fix Lecture 11 optimal batch caption

[LOGO127]

Summary

• Correct the explanation of the red curve on slide 11 of lecture_11.pdf.
• Clarify that the curve connects the minima of parabolic fits to equal-loss points.
• State that, for a fixed model and target loss, the selected batch size is the one requiring the fewest training tokens.

Why

The previous caption described the curve as selecting minimum-loss points at each fixed y-value. This reverses the construction used in MiniCPM: the target loss is held fixed, and token consumption on the y-axis is minimized across batch sizes.

The revised wording follows MiniCPM Section 3.2 and its Appendix A.2.

Fixes #46.

Validation

• Rendered all 58 pages before and after the edit and compared them pixel-by-pixel; only slide 11 changed.
• Rendered slide 11 at high resolution and inspected text alignment, spacing, and clipping.
• Confirmed the PDF remains tagged, contains 58 pages, and preserves the original page size.
• Confirmed the old caption is absent from the searchable text layer and the replacement text is extractable.