PERF: Improve duplicated perf

[sinhrks]

• closes PERF: core/base/IndexOpsMixin duplicated should be changed to use same impl as frame.duplicated #10235
• tests added / passed
• passes git diff upstream/master | flake8 --diff
• whatsnew entry

Index/Series.duplicated now uses dtype-based logic. Also skip algorithm if Index.is_unique is True (expected to be cached in practical situation).

asv:

   before     after       ratio
  [bb6b5e54] [16c6da4f]
-   34.41ms     4.69ms      0.14  algorithms.algorithm.time_int_duplicated
-   53.77ms     5.59ms      0.10  algorithms.algorithm.time_float_duplicated
-   60.22ms    40.20μs      0.00  algorithms.algorithm.time_int_unique_duplicated
SOME BENCHMARKS HAVE CHANGED SIGNIFICANTLY.

NOTE: can create template for htable after #13716.

[jorisvandenbossche]

int -> float

[sinhrks]

Ah thanks. re-submit a bench:)

[codecov-io]

Current coverage is 84.58% (diff: 100%)

Merging #13751 into master will increase coverage by <.01%

@@             master     #13751   diff @@
==========================================
  Files           141        141          
  Lines         51233      51258    +25   
  Methods           0          0          
  Messages          0          0          
  Branches          0          0          
==========================================
+ Hits          43331      43356    +25   
  Misses         7902       7902          
  Partials          0          0          

Powered by Codecov. Last update 2c047d4...12fb5ac

[sinhrks]

Updated the bench.algorithms.algorithm.time_int_unique_duplicated now benchmark under is_unique is once cached.

[jreback]

maybe we can use templates for these (make another issue)