Fix PolyfaceAuxData index blocking flatbuffer format / Reorganize test data

[dassaf4]

Addresses #6583.

TypeScript FlatBuffer serialization of PolyfaceAuxData indices lacked the 0-blocking of the other Polyface index arrays, which led to crashes during mesh processing. Now TypeScript FlatBuffer serialization outputs the correct index blocking, and both TypeScript and native FlatBuffer APIs read both old and new PolyfaceAuxData index formats. The old format is detected using heuristics.

Although there is a similar imodel-native PR, the fixes in each repo are independent of the other.

Also:

• Typescript FlatBuffer and JSON deserialization APIs were incorrectly reading fixed-sized Polyface indices, which have numPerFace > 2 and are only written by the native FlatBuffer API. Move index array reblocking logic to SerializationHelpers for common usage.
• New crossPlatform.test.ts verifies that Typescript API can read native API geometry serializations. (The same test exists in imodel-native to verify the native API can read TypeScript API geometry serializations.) Each test case consists of at least four files: FlatBuffer and JSON serializations of a single geometry, written by both APIs. The test verifies that a given API deserializes each file in a test case into the same geometry. Initial test cases include Polyface variations in numPerFace value, PolyfaceAuxData format, and all combinations thereof.
• IndexedMeshProps was incomplete and used the wrong label for TaggedNumericDataProps.
• Add "Props" interfaces for PolyfaceAuxData and use them in the JSON deserializer.
• Use correct assert(!"foo") idiom
• Split out index validation from IndexedPolyface.terminateFacet so that it can be called once after all facets are added, instead of after each facet. This new index validation caught at least one bug in a mesh created for a PolyfaceQuery test.
• Prefer class members of form foo?: bar over foo: bar | undefined
• PolyfaceData.compress now compresses PolyfaceAuxData if exactly one AuxChannelData is present
• Sphere ctor now validates latitude sweep to avoid common user error. Sphere.createEllipsoid was erroneously capturing inputs.
• Add type predicates to Checker.testDefined and Checker.testUndefined to simplify many unit tests. The linter pointed out the simplifications to the tests.
• GeometryCoreTestIO: add helper functions for tests that perform FlatBuffer/JSON <-> file/GeometryQuery conversions
• Move all unit test input data into one "data" directory, similar to native geomlibs. Hardcoded data paths in tests are updated.
• Improve documentation for PolyfaceAuxData. In particular emphasize that the indices must have the same blocking as the other Polyface index arrays.

[saeeedtorabi]

I only reviewed the last commit where you extended getPointCurveClosestApproachXYNewton to support LineString. Looks good!

[pmconne]

Approving but consider putting the massive code reorganization changes into a separate PR or at the very least changing the PR title to describe what's actually being changed.

[dassaf4]

Approving but consider putting the massive code reorganization changes into a separate PR or at the very least changing the PR title to describe what's actually being changed.

Thanks. Yeah, this one got out of control. Will definitely do next time.