ENH: sparse: speedup LIL matrix slicing (was #3338)

[perimosocordiae]

This is a continuation of @pv's LIL enhancements PR (#3338). I rebased on master and fixed up the benchmarks for sparse slicing to include the LIL format:

BENCHMARK                                    BEFORE      AFTER     FACTOR
...se.NullSlice.time_3_rows(0.01, 'lil')   229.53μs    60.48μs   0.26350105x
...se.NullSlice.time_3_rows(0.05, 'lil')   285.01μs    61.80μs   0.21684491x
...ullSlice.time_10000_rows(0.05, 'lil')   554.35ms    18.44ms   0.03326798x
...ullSlice.time_10000_rows(0.01, 'lil')   431.39ms     7.70ms   0.01785378x

[codecov-io]

@@            master   #5789   diff @@
======================================
  Files          235     235       
  Stmts        43219   43257    +38
  Branches      8143    8152     +9
  Methods          0       0       
======================================
+ Hit          33601   33639    +38
  Partial       2591    2591       
  Missed        7027    7027       

Review entire Coverage Diff as of a6a513f

Powered by Codecov. Updated on successful CI builds.

[larsmans]

@rgommers and @jnothman already gave a 👍 at #3338 and I presume @pv is also still happy with this code; the benchmark code should be uncontroversial. Merging as the diff looks fine.

[pv]

I have no memory of this code, but looks OK.