Improve property access performance

[jmerle]

What does this implement/fix? Explain your changes.

This PR improves property access performance considerably:

• Replaced GetValue/SetValue calls for property/field access with FastMember's TypeAccessor when possible. This is considerably faster than reflection, but not supported in all cases. In cases where it's not supported we fall back to normal reflection.
• Replaced a ConcurrentDictionary in PyObjectConversions with a normal dictionary and added a Count == 0 check before calling TryGetValue on it.
• Wrapped code related to tracking managed objects in an if so it only gets called when it's actually useful.

This PR adds 36 tests covering Python code getting/setting C# properties/fields. The changes pass these tests and all of LEAN's regression tests.

Unrelated changes:

• Doubled the assembly loading timeout to fix #50.
• Bumped the version to 2.0.2.

IndicatorRibbonBenchmark benchmarks (we currently use a Python.NET debug build in production):

LEAN debug master // Python.NET debug master:
Debug: Algorithm Id:(IndicatorRibbonBenchmark) completed in 149,20 seconds at 11k data points per second. Processing total of 1.578.463 data points.
Debug: Algorithm Id:(IndicatorRibbonBenchmark) completed in 146,29 seconds at 11k data points per second. Processing total of 1.578.463 data points.
Debug: Algorithm Id:(IndicatorRibbonBenchmark) completed in 146,48 seconds at 11k data points per second. Processing total of 1.578.463 data points.
Debug: Algorithm Id:(IndicatorRibbonBenchmark) completed in 148,64 seconds at 11k data points per second. Processing total of 1.578.463 data points.
Debug: Algorithm Id:(IndicatorRibbonBenchmark) completed in 147,71 seconds at 11k data points per second. Processing total of 1.578.463 data points.
Average time: 147,664 seconds

LEAN debug master // Python.NET debug PR:
Debug: Algorithm Id:(IndicatorRibbonBenchmark) completed in 115,87 seconds at 14k data points per second. Processing total of 1.578.463 data points.
Debug: Algorithm Id:(IndicatorRibbonBenchmark) completed in 115,61 seconds at 14k data points per second. Processing total of 1.578.463 data points.
Debug: Algorithm Id:(IndicatorRibbonBenchmark) completed in 120,06 seconds at 13k data points per second. Processing total of 1.578.463 data points.
Debug: Algorithm Id:(IndicatorRibbonBenchmark) completed in 119,92 seconds at 13k data points per second. Processing total of 1.578.463 data points.
Debug: Algorithm Id:(IndicatorRibbonBenchmark) completed in 118,67 seconds at 13k data points per second. Processing total of 1.578.463 data points.
Average time: 118,026 seconds (-20.1% against Python.NET debug master)

LEAN debug master // Python.NET release master:
Debug: Algorithm Id:(IndicatorRibbonBenchmark) completed in 89,52 seconds at 18k data points per second. Processing total of 1.578.463 data points.
Debug: Algorithm Id:(IndicatorRibbonBenchmark) completed in 88,79 seconds at 18k data points per second. Processing total of 1.578.463 data points.
Debug: Algorithm Id:(IndicatorRibbonBenchmark) completed in 88,47 seconds at 18k data points per second. Processing total of 1.578.463 data points.
Average time: 88,93 seconds (-39.8% against Python.NET debug master)

LEAN debug master // Python.NET release PR:
Debug: Algorithm Id:(IndicatorRibbonBenchmark) completed in 65,42 seconds at 24k data points per second. Processing total of 1.578.463 data points.
Debug: Algorithm Id:(IndicatorRibbonBenchmark) completed in 66,76 seconds at 24k data points per second. Processing total of 1.578.463 data points.
Debug: Algorithm Id:(IndicatorRibbonBenchmark) completed in 66,83 seconds at 24k data points per second. Processing total of 1.578.463 data points.
Average time: 66,34 seconds (-55.1% against Python.NET debug master, -25.4% against Python.NET release master)

Does this close any currently open issues?

Closes #50.

Any other comments?

No.

Checklist

Check all those that are applicable and complete.

• [x] Make sure to include one or more tests for your change
• [ ] If an enhancement PR, please create docs and at best an example
• [ ] Add yourself to AUTHORS
• [ ] Updated the CHANGELOG

[jmerle]

Summary of the latest changes:

• Replaced FastMember with Fasterflect because it's slightly faster and because LEAN also uses Fasterflect.
• Re-added an if (inst is Exception) that was lost in a previous rebase (https://github.com/QuantConnect/pythonnet/blob/1.0.5.30/src/runtime/clrobject.cs#L33).

New benchmarks:

Debug: Algorithm Id:(IndicatorRibbonBenchmark) completed in 62,48 seconds at 25k data points per second. Processing total of 1.578.463 data points.
Debug: Algorithm Id:(IndicatorRibbonBenchmark) completed in 62,13 seconds at 25k data points per second. Processing total of 1.578.463 data points.
Debug: Algorithm Id:(IndicatorRibbonBenchmark) completed in 63,80 seconds at 25k data points per second. Processing total of 1.578.463 data points.
Average time: 62,80 seconds (-57.8% against Python.NET debug master, -29.4% against Python.NET release master)