Sine we're not using sudo we can migrate to the container-based Travis
build infrastructure [1]. This has the following advantages:

 - Builds start in seconds
 - More available resources
 - Better network capacity, availability and throughput
 - Caching available for open source projects

The builds will still fail due to an `OutOfMemoryError` but that's a
different issue.

 [1] https://docs.travis-ci.com/user/migrating-from-legacy/

reported by Alexander (vraimax2016)

spatial key fix for null geometries

since it breaks all the tests.

this reverts commit 37641c9a

Table alias get lost on joins

Handle LocalTime with nanosecond resolution

TestMemoryUsage would sometimes manage to overrun it's loop bounds and
enter an infinite loop.

TestMultiThread would sometimes timeout if the machine was busy.

TestOutOfMemory could sometimes fail, and would then cause future tests
to fail because it held onto the memory.

FilePathReorderWrites was not deleting it's subsidiary memFS: data.

TestPreparedStatement::testDateTime8 had a typo.

The change to TestMVStore::testCacheSize I'm not sure about.

Recover for "page store" H2   breaks LOBs consistency

Run tests on Travis

Run the tests on Travis so that when a pull request breaks some tests
we notice it immediately.

Add support for Java 8 Date and Time API AKA JSR-310. On the face of it
the existing JDBC types java.sql.Date, java.sql.Time and
java.sql.Timestamp offer conversion methods to and from JSR-310 data
types. Unfortunately java.sql.Timestamp does silent data truncation
when the timestamp falls into a DST transition on the JVM time zone.
Worse still for java.time.OffsetDateTime there is no corresponding JDBC
type. Therefore these conversions have to be implemented. All of this
has to be programmed using reflection so the code compiles and can be
executed on Java 7.

Fix bug in picking the right index for INSERT..ON DUPLICATE KEY UPDATE when there are both UNIQUE and PRIMARY KEY constraints.