A symbolic cubic's linear factors are found before the partial fractions

[Rafael-SOWNet]

Part of #718.

Partial fractions over a symbolic polynomial need its linear factors, and nothing looked for them when the coefficients had symbols in them. WithSymbolicLinearFactorsFound now tries a polynomial of degree three or more with symbols in it, when its leading and constant coefficients are single terms. The candidates are linears u ± v x, with u a divisor of the constant term and v one of the leading coefficient, as for rational roots over the integers. The ones that divide exactly are written apart. If the numerator holds one of them too, it is cancelled from both sides, once per power, before the partial fractions run. Without that cancellation, 1/(a sec x + b tan x)^4 went from a 3-second decline to a timeout.

2.5.0 and master 3c86dda7 declined all six of these rows:

integrand	Rubi	length	time
1/((a + b x)(c + d x)(h + f x)), multiplied out	1.3.1:25	288	0.4 s
the same, squared	1.3.1:26	3,047	1.3 s
the same, cubed	1.3.1:27	10,662	2.4 s
sec(c + d x)^6/(a cos(c + d x) + b sin(c + d x))	4.7.2:150	5,573	0.1 s
sec(c + d x)^3/(a cos(c + d x) + b sin(c + d x))^4	4.7.2:177	9,322	1.5 s
1/(a sec x + b tan x)^4	4.7.7:354	6,548	0.3 s

Lengths are characters of Stringize(). All six ran in one process, in this order.

Rubi corpus, both arms built from 88354428, fix against master:

• Files 0–8, plus a 14,817-row pocket: the pocket goes from 14,619 to 14,625 solved. Every file has the same counts in both arms, and both arms have 0 wrong.
• Rows that moved: the six above. A rerun of only those rows gives 6/6 solved on this branch, graded by differentiating back. On master the same rerun gives 5 declined and 1 timeout.
• Allocation gate: passed.
• Unit suite: 15,195 passed, 0 failed, 13 skipped (net10.0).

BREAKING-CHANGES.md has an entry, A symbolic cubic's linear factors are found before the partial fractions. SymbolicLinearFactorsOfACubicIntegralTest integrates the cubic and its square, and checks each answer differentiates back at sampled points.

🤖 Generated with Claude Code

https://claude.ai/code/session_012sonx8iAspMiwRwokT1Ura