Improve Quantopian compatibility around get_open_orders()

Also unpleasant warning from IPython package is removed.

Retry from temporary error

This should help the process be tolerant in the case of temporary error caused by network, bad data etc. It is sometimes important to be resilient in live trading. It can be disabled by -r false

Comply the NaN handling in price data

When price data does not have particular time window, it was filling by the previous values (method='ffill'). This was not consistent with Quantopian and also causing the aggregated volume value to be bogus.

Critical bug fix around process restart

• Fix wrong hash value used for equality comparison for Asset objects, causing restarted process to get confused with Asset key in dict #27

Minor fix for alpaca backend data masking

It was found and corrected that the alpaca backend get_bars() may return NaN for a minute bucket if there is no data for this specific window, but the zipline spec is presumably to fill by the last value.

Support shell subcommand

With pylivetrader shell <algofile>.py command, you can get into IPython interactive shell with the algo file module namespace. This allows quicker experiments while developing algo.

Experimental smoke testing support

This version includes "smoke" testing for algo with synthetic dummy price data and execute the code to catch potential typos and easy mistake.