Drop the cost percentage claims from the GLM-5.3 post - #4483

The GLM-5.3 post led with "under a third" and "at least 70% less" against buying tokens. Both hold only at the $4.00 per chip-hour rate it uses, which partners consider below today's market rates.

• The description, intro, and conclusion now say that self-hosting costs less than buying the same tokens. The conclusion adds that this holds on GB200 NVL72 without patching the engine or writing kernels.
• The $12.77 and $4.00 figures stay only next to their SemiAnalysis source: the paragraph under the intro, the cost chart, the note, and the limitation.
• The thumbnail no longer has the "Self-hosting 70% cheaper" line (dstackai/static-assets@5e38454, same URL).

Stacked on #4478.

🤖 Generated with Claude Code