PERF: enhance HDFStore Table writing performance

[jreback]

Took out all of the cython code, turns out you can assign directly to numpy recarrays per
column if you put in the correct shape yourself

Invoked with :
--ncalls: 5
--repeats: 5
-------------------------------------------------------------------------------
Test name                                    | head[ms] | base[ms] |  ratio   |
-------------------------------------------------------------------------------
write_store_table_mixed                      |  45.6208 | 121.5352 |   0.3754 |
write_store_table_panel                      |  59.4140 |  99.5456 |   0.5969 |
write_store_table                            |  40.2264 |  65.8978 |   0.6104 |
write_store_table_dc                         | 143.3910 | 174.9424 |   0.8196 |
read_store_table_wide                        |  16.8930 |  17.5310 |   0.9636 |
query_store_table_wide                       |   9.6692 |   9.7564 |   0.9911 |
write_store_table_wide                       | 117.8906 | 118.8232 |   0.9922 |
read_store_table_mixed                       |   4.7604 |   4.7478 |   1.0026 |
query_store_table                            |   4.2646 |   4.2320 |   1.0077 |
read_store_table                             |   2.3678 |   2.3178 |   1.0216 |
read_store_table_panel                       |  20.9388 |  20.3526 |   1.0288 |
-------------------------------------------------------------------------------
Test name                                    | head[ms] | base[ms] |  ratio   |
-------------------------------------------------------------------------------

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

Target [d11041d] : PERF: enhance HDFStore Table writing performance
Base   [aca1a42] : Merge pull request #4437 from cpcloud/fix-astype-calls

[dragoljub]

Jeff this looks fantastic for writing tables. Does this enhancement also apply when appending to an existing table?

I cant wait to give this a shot once 0.13 is out. 👍

[jreback]

http://pandas.pydata.org/pandas-build/dev/

there are are binaries out (iirc you use windows)

this works for appending / writing (it's the same operation )