Add support for timers in Portable Ray Runner

[pabloem]

The prototype outlined in https://github.com/ray-project/ray_beam_runner/pull/10 supports batch pipelines that read / write data.

For streaming workflows, the state and timer primitives exist, which are used to manage long-term persisted state, and managing event time / processing time in a streaming pipeline (see this blog post for more information).

In fact, windows and triggers (higher-level concepts) for Beam can be built using state and timers.

Timers are passed in a similar way that data is passed in Beam (serialized and sent to the runner by the worker), so we only need to process them and schedule them for execution.

Code pounters

• The FnApiRunner:
  • Receiving timers from the SDK worker: https://github.com/apache/beam/blob/1f8ae6539d70ffd85d64b6e49c103309476756b2/sdks/python/apache_beam/runners/portability/fn_api_runner/fn_runner.py#L1296-L1304
  • Scheduling timers for execution: https://github.com/apache/beam/blob/e42027b4cc34c6501f263d4ae641de2bd03887b1/sdks/python/apache_beam/runners/portability/fn_api_runner/fn_runner.py#L659-L684
• The Ray Runner
  • Receiving data from the SDK worker: https://github.com/ray-project/ray_beam_runner/blob/master/ray_beam_runner/portability/execution.py#L124-L129

[pabloem]

Useful literature: https://s.apache.org/beam-fn-api