Velocity and acceleration are computed using automatic differentiation.

Conflicts:
	src/main/java/org/orekit/errors/OrekitMessages.java
	src/main/resources/assets/org/orekit/localization/OrekitMessages_de.utf8
	src/main/resources/assets/org/orekit/localization/OrekitMessages_el.utf8
	src/main/resources/assets/org/orekit/localization/OrekitMessages_en.utf8
	src/main/resources/assets/org/orekit/localization/OrekitMessages_es.utf8
	src/main/resources/assets/org/orekit/localization/OrekitMessages_fr.utf8
	src/main/resources/assets/org/orekit/localization/OrekitMessages_gl.utf8
	src/main/resources/assets/org/orekit/localization/OrekitMessages_it.utf8
	src/main/resources/assets/org/orekit/localization/OrekitMessages_no.utf8
	src/main/resources/assets/org/orekit/localization/OrekitMessages_ro.utf8
	src/test/java/org/orekit/errors/OrekitMessagesTest.java

Merge branch 'orbit-determination' of ssh://www.orekit.org/orekit-labs/orekit-od into orbit-determination

Checks for small bracketing intervals.

Ported a large suite of event tests from hipparchus ODE integrators to Orekit
Propagators. Created several sub tests to run these with different analytical
and numerical Propagators. Also included recent updates to to the Hipparchus
event finding algorithm.

At this time only the DP853 integrator seems to work reliably with events. The
Adams integrators have trouble with resets and the GBS integrator gets stuck in
an infinite loop. At this point I'm not sure if these are bugs per se or
limitations that need to be documented.

Conflicts:
	src/main/java/org/orekit/propagation/integration/IntegratedEphemeris.java
	src/site/markdown/tutorial/integration-in-Matlab.md
	src/site/xdoc/changes.xml

This improves robustness in corner cases, typically when different
event detectors triggers at very close times and one of them
resets the state such that it affects the other detectors.

The setDate/GetInterpolatedState() pair has been replaced by a single
getter.

The caching did "save" computation of g function values not only
when the same time was reused, but also for times one ulp away.
This broke events detection as when we are very close to a zero,
a one ulp difference in time changes the sign, by definition of
the crossing.