Spanner Performance Improvements

[blowmage]

We have received feedback that Spanner's serialization to a Ruby Hash is not as performant as expected. While there are micro-optimizations that can be made, the largest improvement in performance can be made by optionally skipping the check for duplicate names. Spanner data can contain duplicate names, or no names, which makes representing it in a Ruby Hash difficult. Consider the following example:

hash = { foo: :bar, foo: :baz } #=> {:foo=>:baz}

The former value of :bar was dropped in favor of the latter value of :baz. To ensure that values are not dropped, Spanner data will raise DuplicateNameError when serializing to a Ruby Hash. This PR adds the optional skip_dup_check argument that will avoid the check for users that want the most performance and are willing to manage the risk of losing values.

require "google/cloud/spanner"

spanner = Google::Cloud::Spanner.new

db = spanner.client "my-instance", "my-database"

dup_type = db.fields [[:x, :INT64], [:x, :STRING], [:x, :BOOL]]
dup_data = dup_type.struct [42, nil, false]

dup_data.to_h # raises Google::Cloud::Spanner::DuplicateNameError

dup_data.to_h skip_dup_check: true #=> {:x=>false}

Here are the performance numbers for master, the performance tweaks, and the performance tweaks with skipping the duplicate name check:

                        user     system      total        real
master (GC on)      0.260000   0.000000   0.260000 (  0.269273)
perf only (GC on)   0.240000   0.000000   0.240000 (  0.245119)
perf skip (GC on)   0.140000   0.000000   0.140000 (  0.148022)
master (GC off)     0.220000   0.020000   0.240000 (  0.232874)
perf only (GC off)  0.200000   0.010000   0.210000 (  0.216652)
perf skip (GC off)  0.130000   0.000000   0.130000 (  0.141780)