Track header files in CMake targets for build impact analysis

[pratikpugalia]

Description

Motivation

Build Impact Analysis

Today, there is no reliable way to answer: "If I change Foo.h, which CMake targets need to rebuild?"

CMake's File API (codemodel-v2) reports source files associated with each target. However, headers
added via install(FILES ...) or glob-based helpers are invisible to the File API — they exist on disk but have no target association in the build graph.

By using FILE_SET HEADERS (CMake 3.23+), every header becomes a first-class source of its target, queryable through the File API.

Selective Builds

With accurate header-to-target mapping, we can build selective build tooling:

1. A PR modifies velox/core/PlanNode.h
2. Query the File API → PlanNode.h belongs to target velox_core
3. Build and test only velox_core and its dependents, instead of the entire project

This significantly reduces CI time for targeted changes.

Approach

1. HEADERS keyword in velox_add_library() — production targets declare their headers explicitly, wired to FILE_SET HEADERS internally.

2. velox_add_test_headers() — lightweight helper for test/benchmark/fuzzer targets using raw add_library()/add_executable().

3. CI enforcement — a check-header-ownership pre-commit hook verifies every .h file under velox/ is referenced in a CMakeLists.txt. New headers without ownership fail CI.

4. Compatibility — VELOX_MONO_LIBRARY mode fully supported. Existing velox_install_library_headers() kept as fallback. No changes to build output or installed artifacts.

Scope

• ~1,287 headers tracked across ~150 CMakeLists.txt files
• ~65 new INTERFACE targets mirroring BUCK's header-only structure

Future Work

• Build selective build tooling consuming the File API data

Related PR: #16897

cc @kgpai @czentgr @majetideepak

[srsuryadev]

@pratikpugalia