Keep the sign of a half-odd power of a negative multiple through the long division

[Rafael-SOWNet]

The polynomial reader took (c x)^r apart as c^r x^r for any constant c and a power r that is not whole. That holds for c > 0. For c < 0, c x is positive where x < 0, and c^r and x^r are both imaginary there, so the answer comes out with the wrong sign. Such a power now comes apart over a positive number only. The limits at +oo read the variable as positive and tell the reader so.

Wrong answers in 2.5.0, each the negative of the right value for every x < 0, where the integrand is real:

	2.5.0	Now
sqrt(-x)*(x+1)/x, Integrate	2i * (sqrt(x)) ^ 3 / 3 + 2i * sqrt(x) + C	(sqrt(-x) * x / (3/2) + sqrt(-x) / (1/2) provided not x = 0) + C
sqrt(-x)*(x+1)/x, Simplify	i * (sqrt(x) + x ^ (-1/2))	unchanged
(-3*x)^(5/2)*(x+1)/x, Integrate	(-3)^(5/2) x^(7/2)/(7/2) + …	((-3) x)^(5/2) x/(7/2) + …

Answered as a result, with b under the power and right for either sign of b:

• (b x)^(5/2) (x^2 - 1)/x;
• Rubi's 4.3.0 (b sec(x))^m tan(x)^3 and cot(x)^3 (b csc(x))^m at a half-odd m.

Measured on 8a1ef454, branch against master:

Run	Problems	Master	Branch
Pocket: the files that reach the reader	6,815	6,758 solved	6,757 solved
Family 1, 40 a file	1,381	1,362	1,362
Family 0, whole	1,814	1,789	1,789
Families 2 to 8, sampled	2,410	2,362	2,361

Two problems moved. Both time out on both builds when re-run alone, so they are load noise. No run had a wrong answer. Gate and suite (15,618) pass.

AHalfOddPowerOfANegativeMultipleIntegralTest checks both signs of b and of the variable. BREAKING-CHANGES.md has the entry.

Part of #233.

🤖 Generated with Claude Code

https://claude.ai/code/session_012sonx8iAspMiwRwokT1Ura