ruby: Improve performance of Google::Protobuf::Timestamp#to_time (#6360)

This changes to_time to use Ruby's built in Time.at with nanos support
rather than calculating a float and passing it to Time.at. The new
version runs about 3 times faster than the original version and
allocates fewer objects.

Warming up --------------------------------------
    protobuf#to_time    57.296k i/100ms
      faster#to_time   133.229k i/100ms
Calculating -------------------------------------
    protobuf#to_time    635.361k (± 2.1%) i/s -      3.209M in   5.052169s
      faster#to_time      1.873M (± 3.3%) i/s -      9.459M in   5.055169s

Comparison:
      faster#to_time:  1873368.8 i/s
    protobuf#to_time:   635361.4 i/s - 2.95x  slower

Calculating -------------------------------------
    protobuf#to_time   326.000  memsize (   126.000  retained)
                         7.000  objects (     2.000  retained)
                         0.000  strings (     0.000  retained)
      faster#to_time    86.000  memsize (     0.000  retained)
                         1.000  objects (     0.000  retained)
                         0.000  strings (     0.000  retained)

Comparison:
      faster#to_time:         86 allocated
    protobuf#to_time:        326 allocated - 3.79x more

ruby/lib/google/protobuf/well_known_types.rb

@@ -72,8 +72,14 @@ module Google
     end
 
     Timestamp.class_eval do
-      def to_time
-        Time.at(self.to_f)
+      if RUBY_VERSION < "2.5"
+        def to_time
+          Time.at(self.to_f)
+        end
+      else
+        def to_time
+          Time.at(seconds, nanos, :nanosecond)
+        end
       end
 
       def from_time(time)