BUG: Report a missing typed getter as an assumption

[oleksandrlazarenko-pi]

Closes #584

What this fixes

The Nightly Pipeline failed on nine consecutive nights, 2026-09-02 through
2026-09-10, on every platform/JDK job of every open nightly PR. One test failure
was responsible for all of them:

ValueHashTests.Values_Hash_TypedGetters:149 The properties IsVisible,
IsVisibleJavaScript, HasWebDriver, HasWebDriverJavaScript, IsHeadlessare missing
getters in the DeviceData class.

Why it happened

ValueTests.typedGetters called fail(...) when the data file exposed a property
with no strongly typed getter. The two inputs to that check do not publish in step:

• the data file gained the five properties around 2026-09-01;
• DeviceData/DeviceDataBase are generated by ci/generate-accessors.ps1 →
  tools/PropertyGenerator, which reads the live 51Degrees metadata service.

The metadata service did not publish them until 2026-09-10, so Nightly Data File Change logged "No property changes, so not creating a pull request" every night
from the 2nd to the 9th, and only opened #583 on the 10th. For eight days there was
no change to this repository that could have made the assertion pass.

The change

A missing typed getter is reported as a JUnit assumption failure, so the test is
recorded as skipped with the full diagnostic instead of breaking the build. This is
already the house idiom for conditions outside the repository's control
(MetaDataHashTests, EvidenceKeysTests, EngineTests), and MavenRunListener
surfaces it as Assumption Failed at WARN.

Still hard failures, unchanged:

• a getter that returns null for an available property;
• a getter that does not throw PropertyMissingException for an unavailable one;
• no property having a typed getter at all — the accessors are then broken or
  absent rather than behind, which is this repository's defect.

The reporting is now shared by the on-premise and cloud copies of ValueTests, so
the two cannot drift on this point. The scanning loop above it is still duplicated.

Also fixes the missing space that produced IsHeadlessare missing getters, which ran
the last property name into the sentence in both copies.

Tests

New MissingGettersReportTest (6 tests) in device-detection.shared. Four of them
fail against the pre-fix behaviour and pass with the fix, covering both defects: the
hard failure, and the missing space.

mvn -pl device-detection.shared test — 9 tests, 0 failures.
mvn -pl device-detection.cloud test-compile — success.

Known limitation

Losing a single getter while the rest survive is no longer caught. Detecting that
needs a checked-in list of the getters expected to exist; it is called out in the
javadoc rather than left implicit.

Not fixed here

Nightly Publish's Test stage only exercises the examples against packaged
artifacts and never runs ValueHashTests, which is why main looked green and
packages kept publishing for eight days while the repository's own suite was
failing. That is worth a separate issue.