Add ifds runner for ets

[MForest7]

Add ets-specific IFDS runner. The purpose is to support dynamic calls and (maybe) def-use chains, both need ifds modifications. Also get some performance increase (split path edges storage).

Performance before:

project	stmts	min time	max time	avg time	median time	new inferred types (%)
project1	9914	2.7s	3.2s	2.9s	2.9s	37
project2	26679	3.4s	4.2s	3.7s	3.5s	15
project3	4272	1.2s	1.3s	1.3s	1.3s	29
project4	4584	0.8s	0.9s	0.8s	0.8s	37
project5	121091	70.8s	80.3s	76.9s	78.7s	36
project6	28548	6.6s	6.8s	6.7s	6.7s	38
project7	5162	0.7s	1.1s	0.9s	0.9s	29

After:

project	stmts	min time	max time	avg time	median time	new inferred types (%)
project1	9914	1.1s	1.6s	1.4s	1.4s	36.6
project2	26679	1.2s	1.3s	1.2s	1.2s	15.0
project3	4272	0.5s	0.6s	0.5s	0.5s	29.0
project4	4584	0.3s	0.4s	0.4s	0.4s	36.0
project5	121091	25.8s	37.6s	29.9s	27.2s	35.0
project6	28548	3.1s	3.9s	3.5s	3.3s	36.1
project7	5162	0.5s	0.7s	0.5s	0.5s	28.2

Changes (median time):

project	before	after	decrease
project1	2.9s	1.4s	~50%
project2	3.5s	1.2s	~60%
project3	1.3s	0.5s	~60%
project4	0.8s	0.4s	~50%
project5	78.7s	27.2s	~60%
project6	6.7s	3.3s	~40%
project7	0.9s	0.5s	~40%

[Lipen]

@MForest7 Please, use at most 1 digit after the floating point. 1.6 looks better than 1.654s. I bet you have at most 2 significant digits in your measurements, but overall it is OK to just round all small values (0.1 to 10) to 1 floating digit (like 0.7s or 2.4s), and all larger numbers to integers (30s, 100s).

[Lipen]

OK