Merge branch 'master' of gitlab.aicrowd.com:flatland/flatland into 273_malfunction_loading_and_saving

# Conflicts:
#	examples/simple_example_3.py
#	flatland/envs/rail_env.py
#	flatland/envs/rail_env_utils.py

Malfunction last step missed

Closes #263 and #269

See merge request flatland/flatland!246

faq for flatland

Closes #262

See merge request flatland/flatland!250

Merge branch 'temp_Update_branch' of gitlab.aicrowd.com:flatland/flatland into 273_malfunction_loading_and_saving

will this be to expensive? If yes this can easily be refactored

This resolves issue #273

updated examples

This resolves issue #273

renamed test to correct naming

This resolves issue #273

added test for saving and loading malfunction parameters

This resolves issue #273

fixed tests

This resolves issue #273

replace boolean global numpy array with int numpy array where the agent handle...

Closes #271

See merge request flatland/flatland!255

removed unused example

Closes #264

See merge request flatland/flatland!253

replace boolean global numpy array with int numpy array where the agent handle is set inside the numpy array (#271)

#237 added missing doc string

Closes #237

See merge request flatland/flatland!252

Reintroduced some malfunction parameters but are not used.

At each step we first check for possible malfunctions.
THen we apply them and do all the agent steps
At the end we look at what agents should be fixed. This lead so to:

Current malfunction value says how many step calls are necessary till agent can move again

e.g. malfunction = 2, we have to call step() twice untill the agent can move again.

Code cleanup

At each step we first check for possible malfunctions.
THen we apply them and do all the agent steps
At the end we look at what agents should be fixed. This lead so to:

Current malfunction value says how many step calls are necessary till agent can move again

e.g. malfunction = 2, we have to call step() twice untill the agent can move again.

It now gives the probability that an agent breaks at given time step
given the number of agents and the rate of failures of a single agent.
These number now need to be much larger.... otherwise we will have alot of failures