[numba.md] numba_ex3 asks for n = 100_000_000 but the shared setup and solution use n = 1_000_000 #579

Since #550, the Exercises section opens with a shared setup block that draws u_draws and v_draws at n = 1_000_000 so the samples can be reused by both {ref}speed_ex1 and {ref}numba_ex3.

The statement of numba_ex3 still ends with: "For the size of the Monte Carlo simulation, use something substantial, such as n = 100_000_000." The provided solution then uses the shared 10^6 arrays, so it appears to ignore the exercise's own instruction — and at 10^6 the parallel gain over the serial version is modest, which undercuts the point the exercise is making.

Two possible fixes:

• Soften or drop the n = 100_000_000 sentence in numba_ex3 and point readers at numba_ex_draw_speed, which already does the large-n comparison end-to-end.
• Keep the sentence and add a line to the solution noting that the shared n keeps runtime and memory low, and that scaling up means regenerating the arrays (memory grows at roughly 16n bytes, about 1.6 GB at 10^8).

Origin: flagged by Copilot review on the zh-cn sync PR QuantEcon/lecture-python-programming.zh-cn#70. The translated text mirrors the English source verbatim, so the fix belongs here and will propagate to all editions through translation sync.