Improve performance

[halhen]

Following your request at #14 (comment), I took a stab at speeding generateCombinationsImpl up a bit. My results are 3-4X speedup; the majority of the time is now spent paste():ing combination names.

Not an earth-shattering difference, but also not nothing perhaps?

Unit tests pass, but please verify that I've broken no assumptions.

5950X, 128 GB RAM, Fedora Linux 36, R 4.1.3, v1.11.0 / commit 12f306f

Before this PR:

> microbenchmark::microbenchmark(
+     without_store_elems = upsetjs() %>%
+         fromDataFrame(toyset_2, c_type="distinctIntersection", store.elems=FALSE, limit = 40),
+     with_store_elems = upsetjs() %>%
+         fromDataFrame(toyset_2, c_type="distinctIntersection", store.elems=TRUE, limit = 40),
+     times = 10L
+ )
Unit: milliseconds
                expr      min       lq     mean   median       uq      max neval
 without_store_elems 825.4124 1044.876 1047.124 1078.969 1105.553 1126.978    10
    with_store_elems 943.7230 1031.711 1047.950 1058.206 1082.389 1096.377    10

With this PR:

> microbenchmark::microbenchmark(
+     without_store_elems = upsetjs() %>%
+         fromDataFrame(toyset_2, c_type="distinctIntersection", store.elems=FALSE, limit = 40),
+     with_store_elems = upsetjs() %>%
+         fromDataFrame(toyset_2, c_type="distinctIntersection", store.elems=TRUE, limit = 40),
+     times = 10L
+ )
Unit: milliseconds
                expr      min       lq     mean   median       uq      max neval
 without_store_elems 239.3860 243.9979 251.5668 252.5741 258.3372 261.2554    10
    with_store_elems 305.0873 319.4910 342.5158 328.5697 335.9419 472.6493    10