[rllib] Flip sign of A2C, IMPALA entropy coefficient; raise DeprecationWarning if negative

[ericl]

What do these changes do?

Throw an error if we end up penalizing entropy, which is probably unintended. Other options here:

1. break backwards compat by renaming the config
2. always take the absolute value here?

Related issue number

Closes #4369

[ericl]

Upon further thought, we can actually safely flip the sign, by raising DeprecationWarning if it's negative.

This is probably the better long-term solution, but should be reviewed carefully.

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-Perf-Integration-PRB/91/
Test PASSed.

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-Perf-Integration-PRB/94/
Test PASSed.

[AmplabJenkins]

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/12893/
Test FAILed.

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-Perf-Integration-PRB/102/
Test PASSed.

[AmplabJenkins]

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/12912/
Test FAILed.

[ericl]

jenkins retest this please

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-Perf-Integration-PRB/109/
Test PASSed.

[AmplabJenkins]

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/12922/
Test FAILed.

[ericl]

jenkins retest this please

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-Perf-Integration-PRB/126/
Test PASSed.

[AmplabJenkins]

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/12946/
Test FAILed.