Fix NaN when province admin cost is exactly zero

[ruggsea]

update_administrative_efficiency can turn a province's control_ratio into NaN, and that NaN later spreads to tax income, treasuries and market prices.

Cause. When control_ratio is below 0.01, the factor 1/(1.01 - control) - 1 in the admin cost is negative, so (pop * conc + area * 100) * factor + 100 can come out exactly 0. Then desire = prize / mass is inf, and control_scale * desire is 0 * inf = NaN.

Where we hit it. A headless AI-only campaign (seed 97). On 1869-10-13, province 593 (25,234 people, 3,322.6 km², control 0.00972) had mass == 0. The NaN reached the owner's treasury and every market, and the game aborted in 1871.

Fix. On a tick where mass is exactly 0, leave control unchanged. Every other case computes exactly the same values as before.

Testing. On our Linux headless build (single-threaded, fixed seeds):

• The seed 97 campaign now runs the full 100 years with no NaN.
• Three other seeds produce byte-identical output with and without the change over 100 years (monthly nation, province and price dumps, plus the war log).
• This branch builds on current main.

[ineveraskedforthis]

I would recommend adjusting calculation of admin cost instead to avoid negative numbers by definition.

[ruggsea]

Done, thanks: the skip is gone and the factor is clamped with std::max/ve::max(0.f, ...) in both versions of the admin cost, so the cost is always >= 100.

Checked it on headless AI campaigns: the seed that originally went NaN (1869) and crashed (1871) now runs through 1872 with no NaN in prices or treasuries. 20 seeds x 100 years with vs without the change show no difference in world population, industrial score, treasury, prices, nations alive/civilized or wars started (all Mann-Whitney p > 0.06), and none of the 40 campaigns has a NaN.