BLD: test_perf.py, add --base-pickle --target-pickle options to test_perf

[ghost]

Until now test_perf either generated a single commit report for HEAD,
or fell back to vbench to compare two commits. with this change, can now
compare results between saved results of test_perf -H invocations.

Flow:

• Use build_cache (cdev from Tips and tricks for pandas devs #3156) to jump to desired commit using build_cache
• checkout current vb_suite from upstream/master
• use test_perf -H -d <commit>-foo.pickle to save timings to file, for
  target and base commits
• Use test_perf options --base-pickle, --target-pickle to generate comparison report.

cc @jreback

λ ./test_perf.sh --base-pickle 31ecaa9-0.10.1.pickle --target-pickle f9eea30-0.11.0.pickle
This script compares the performance of two commits.
Make sure the python 'vbench' library is installed.

Setting the BUILD_CACHE_DIR env var to a temp directory will
potentially speed up subsequent runs.

base_commit not specified, Assuming base_pickle is named <commit>-foo.*
target_commit not specified, Assuming target_pickle is named <commit>-foo.*
f9eea30
Verifying specified commits exist in repo...
31ecaa9: RLS: set released to true
f9eea30: RLS: Version 0.11
*** 
Invoked with :
--ncalls: 3
--repeats: 3


-------------------------------------------------------------------------------
Test name                                    | head[ms] | base[ms] |  ratio   |
-------------------------------------------------------------------------------
frame_reindex_columns                        |   0.3260 |   0.2900 |   1.1241 |
-------------------------------------------------------------------------------
Test name                                    | head[ms] | base[ms] |  ratio   |
-------------------------------------------------------------------------------

Ratio < 1.0 means the target commit is faster then the baseline.
Seed used: 1234

Target [f9eea30] : RLS: Version 0.11
Base   [31ecaa9] : RLS: set released to true