core: stdcm: cache engineering allowance sequences

[eckter]

We very often test several engineering allowances at the same place, with the same previous edges. This is done when we can get overtaken by more than one train at once at the same place, which is actually very common.

Caching the sequences avoids redundant computations. We still only evaluate what's necessary, and then use soft references to avoid ram issues.

Full bench data, +3% average peak ram, -25% processing time. Run with 6 GB limit.

	time_old	time_per_knode_old	time_new	time_per_knode_new	time_diff
count	88.000000	88.000000	88.000000	88.000000	88.000000
mean	6.955023	0.706642	5.180443	0.648656	-1.774580
std	14.103137	1.221101	9.960951	1.190449	4.376797
min	0.100000	0.038378	0.098000	0.037140	-27.682000
25%	0.920000	0.224366	0.902250	0.163905	-1.260750
50%	2.997000	0.321516	2.577500	0.269849	-0.373500
75%	6.156500	0.511772	5.032000	0.448302	-0.037500
max	99.435000	6.894737	73.056000	6.631579	0.067000

	mb_old	mb_new	mb_diff
count	88.000000	88.000000	88.000000
mean	3257.806818	3350.636364	92.829545
std	963.730168	858.745117	705.095122
min	1333.000000	1333.000000	-1851.000000
25%	2426.500000	2668.750000	-301.250000
50%	3080.000000	3332.500000	50.000000
75%	3960.250000	3917.750000	492.000000
max	5195.000000	5339.000000	2012.000000

[Erashin]

We'll have to check if other parts of the code can implement this. Gj! :)