columnToFieldIndex: Improve performance

[paulquerna-okta]

This PR is in the same spirit as #347

We found that the function columnToFieldIndex was inefficient from a CPU usage POV, but more importantly created ton of objects for the GC.

Analyzing real world usage of gorp in a proprietary API server, we were able to reduce overall memory allocs by ~50% and bytes allocated by ~20% with this patch.

This PR:

• Adds a few tests and a micro-benchmark for columnToFieldIndex
• Takes a dependency on reflectx to implement columnToFieldIndex. reduce call to reflect.Type.FieldByName #347 tries to implement it locally, but reflectx does exactly what we need here, and has built-in caching based on the reflect.Type. We would need to replicate a few things into gorp if the dependency is not acceptable.

Micro-benchmark results:

benchmark                          old ns/op     new ns/op     delta
BenchmarkCcolumnToFieldIndex-8     7183          334           -95.35%

benchmark                          old allocs     new allocs     delta
BenchmarkCcolumnToFieldIndex-8     77             2              -97.40%

benchmark                          old bytes     new bytes     delta
BenchmarkCcolumnToFieldIndex-8     1256          208           -83.44%

[paulquerna-okta]

(note, I think we need some improved tests and changes for embedded structures, but I hope this is still a directionally interesting PR)

[hi-wayne]

sqlx 1.2 reflectx not suppor embedded structs
fyi sqlx1.3 release
jmoiron/sqlx#429