Fix floating-point sentinel misuse in KDTree and SDT

[Caideyipi]

Description

Fix floating-point sentinel initialization

• Initialize the Swinging Door upper and lower bounds with negative and positive infinity. This prevents a first decreasing slope, including one after reset, from immediately crossing the doors.
• Calculate KDTree bounds across every point with infinity sentinels and independent min/max updates. This prevents incorrect subtree pruning when coordinates are negative or the first point contains an extremum.
• Use Double.NaN as the integration-test NULL sentinel so delta-based assertions distinguish NULL from a real 0.0 value.

Verification

• mvn spotless:apply -pl iotdb-core/calc-commons,iotdb-core/node-commons,iotdb-core/datanode
• mvn spotless:apply -pl integration-test -P with-integration-tests
• mvn checkstyle:check -pl iotdb-core/calc-commons,iotdb-core/node-commons,iotdb-core/datanode
• KDTreeUtilTest: 1 test passed through an independent compile/run
• SwingingDoorTrendingFilterTest: 3 tests passed through an independent compile/run

The standard node-commons and datanode Maven test lifecycles are currently blocked before the test phase by unrelated cross-module missing-symbol errors in the local checkout. The changed source and targeted tests were therefore compiled and executed independently.

---

This PR has:

• been self-reviewed.
• added or modified unit tests to cover the affected code paths.

---

Key changed/added classes

• KDTreeUtil
• SwingingDoorTrendingFilter
• TestConstant