Complete Cabal build with deterministic parallel compilation

[nanavati]

Superseded by #17, which contains both logical commits.

All eight Haskell tools and the three test runners now build through Cabal. Make installs those binaries and builds the native/runtime pieces serially; Cabal owns Haskell dependency scheduling and parallelism.

Generated sources live in each component's build directory. A local Warmup module loads dependency interfaces in a stable order before compilation, and GHC 9.14.1 or newer uses object determinism and a shared job semaphore. Older supported GHC versions build serially by default. Build-affecting environment aliases become tracked Cabal flags, and native source edits trigger the relevant native rebuilds.

CI adds a GHC 9.14.1 determinism check comparing clean builds with 1, 16, 64, 64, and 64 workers, followed by a source edit/revert cycle. It hashes objects, interfaces, libraries, all tools, and test runners. The existing version matrix remains serial below 9.14.1.

This is based on the head of B-Lang-org#1084 (f092cfe34ad9851d67b963aea4c4c61926999ea1). The complete change, including this commit and the compiler-library split, is reviewed in #17.

Validation on GHC 9.14.1 / Cabal 3.16.1.0:

• Built all eight tools and all three test runners; completed Make installation and license generation.
• Smoke, utility, and explicitly enabled VCD checks: 336 passes and two expected failures.
• Verified generated-source isolation, native edit/revert rebuilds, and tracked version-policy flags.
• Created a source archive and tested native builds and Cabal configuration outside Git.
• Build determinism: all 1,028 products match across clean builds with 1, 16, 64, 64, and 64 workers. The edit/revert check passes, including each test runner's interfaces.

The reported Git revision is captured on component build. Cabal 3.16 can retain that stamp after a Git-only change until the component is rebuilt; INSTALL.md documents this limitation. The determinism check covers one checkout and toolchain, not reproducibility across paths or machines.