tests: Lower default of iters to 4 (but keep higher on CI)

[real-or-random]

See discussion at https://gnusha.org/secp256k1/2024-07-01.log, also mentioned in #1549.

[jonasnick]

@sipa said the following in the discussion:

< sipa> i don't think that running at more than 1 has ever actually contributed to a bug being found :p

I don't agree. For example, there are tests that should draw random group elements with an even Y coordinate in one iteration and an odd one in another to test both code paths. Default iteration number of 4 will make 1 in 16 such tests useless.

[sipa]

@jonasnick I don't mean to imply that running with more than 1 iteration doesn't add anything; it clearly adds cases that aren't performed otherwise. I literally meant that I don't think we have ever found an actual bug through those.

[jonasnick]

Should have spelled this out more clearly: I have found bugs during development in code that was only correct if the Y coordinates of random group elements had a specific parity :/

[jonasnick]

See also #1036