Pylivetrader

In this release, we strengthen the integration between the two packages: pylivetrater and pipeline-live to work together to have daily screening capabilities.
* you can now use `attach_pipeline` and `pipeline_output` to use pipeline live inside pylivetrader
* created new example code to get the user started:
* simple MACD example (updated)
* simple pipeline-live usage example
* added smoke scripts to example algos. check the READMEs
* added argument for pipeline output hook to Algorithm()
* udpated the minimum Alpaca SDK to 0.46

Fixes
* don't pickle your pipeline when storing the state file
* fix the smoke tool to work with mocked pipeline output
* small fixes to smoke backend to be able to work with pipeline-live mock
* we change the yaml config loading to use the safe loading (security fix)

In this release, we fix some of the issues that occured since the previous release.
* update the example code int he
* don't yield the BAR event at session close (when trading day is over)
* make sure SESSION_END is yielded once per trading day
* fix the shell (set datetime in the data object)
* call initialize(context) in the shell for the user (user don't need to do that anymore)
* add shell commands description to the readme
* update the bottleneck package
* remove the \_\_setattr\_\_ protection from the Algorithm() class. not needed and breaks the smoke tests.

In this release, we upgrade the Alpaca SDK to 0.38

This release upgrades the version of the Alpaca SDK, a major dependency for pylivetrader's Alpaca backend. Due to the update to the websockets library, this means that *support for Python 3.5 is ended.* As of this version, the only supported version is Python 3.6. The lack of explicit support for Python 3.7 at the moment is due to a compatibility issue with some of the Zipline dependencies when building. We'll continue exploring options to restore Python 3.7 support, but for now, *Python 3.6 is the recommended version for Pylivetrader.*

Additionally, a fix was made to the handling of timestamps such that they should now once again work with Pipeline live's USEquityPricing. A fix was also made to smoke test to support algorithms using the all_orders method.

A user found that calling` data.current(symbol('UJB'))` was consistently resulting in errors because of a lack of data on the symbol from Polygon. Now, an empty dataframe will be returned rather than an error. Also, iexfinance released an update which breaks pipeline-live's previous version, so I'm updating the dockerhub dependency to point to its current version.