Change architecture to match polly input test limit

[laardee]

"The size of the input text can be up to 1500 billed characters (3000 total characters). SSML tags are not counted as billed characters." http://docs.aws.amazon.com/polly/latest/dg/limits.html

Possible solution would be to preprosess text to chunks before saving to S3 bucket, then convert function would send those chunks separaterly and concatenate mp3 files to one with ffmpeg.

In this case it would be better to write podcast xml after audio files are ready, so maybe DynamoDB shoud be added what contains details.