Powerset::fold specialization

[Philippe-Cholet]

Related to #755

Rank	Description	powerset_0_fold (µs)	powerset_1_fold (µs)	powerset_2_fold (µs)	powerset_4_fold (µs)	powerset_8_fold (µs)	powerset_12_fold (µs)
3	no specialization	[321.26 322.12 323.18]	[395.73 396.62 397.61]	[353.65 355.59 357.87]	[333.72 334.66 335.72]	[321.30 321.79 322.31]	[318.88 319.67 320.71]
1 wins	currently	[312.58 312.94 313.29]	[375.18 375.65 376.12]	[330.77 331.83 332.96]	[298.66 299.17 299.72]	[289.46 289.81 290.18]	[289.31 289.68 290.16]
2	With (k+1..=n).fold	[320.78 322.00 323.43]	[383.65 384.39 385.30]	[349.08 350.11 351.12]	[315.60 316.55 317.63]	[304.82 305.70 306.59]	[296.88 297.81 299.03]

So it's 10% faster than without specialization for powerset_{4,8,12}_fold (and 3% to 7% for n=0,1,2).
And (it.k() + 1..=it.n()).fold makes it (intermediary) slower (so it's not pushed here).

Once Combinations::fold is implemented, it should be even faster.

EDIT: Sorry, the table does not render well.

[jswrenn]

Thanks! Can you add a specializations test? See the comment here for guidance: #764 (review)