Hello!

I noticed that each Date#to_bson call creates new instance of Time - it's not very problematic, but is noticeable when serializing big amount of Dates.

I'm attaching small benchmark with proposed improvement (use #jd method and calculate UNIX timestamp * 1000 right there in to_bson method).

For 1 million dates, it's 0 allocations vs almost 100MB worth of allocations, also this way is about 80% faster.

Is that OK idea or would that cause some problems?

Best,
Michał