Make zig-physics build; 0 tests were running out of 640

[gHashTag]

The state this repository was in

87 source files, 640 test blocks, no workflow of any kind, and a
manifest that was not ZON:

build.zig.zon:1:7: error: expected ';' after statement
{.name: "zig-physics", version: "0.1.0", dependencies: {}}
      ^

That is JSON. zig build stopped at line 1, column 7, so no compiler had
ever reached a single one of the 87 files.

What was underneath

Fixed in order, each one revealed by fixing the one before it:

1. Manifest — rewritten as ZON, with the .fingerprint Zig requires
   (it proposed the value itself).
2. b.build_root_path — no such field; it is b.build_root.path. The
   line sat inside if (…) |x| else |_| {}, so had it compiled, a failure
   would have been swallowed silently.
3. Two unused locals in build.zig — both addModule results discarded.
4. addTest — took .root_source_file; since 0.15 it takes
   .root_module.
5. The existence guard — nine modules were declared inside an
   inline for that checked whether each file existed and silently skipped
   the ones that did not. All nine exist, so it protected nothing; it only
   guaranteed that a module going missing would stop being exported without
   anybody being told. Declared directly instead.
6. Fossil syntax in quantum/e8_integration.zig — extern enum(c_int),
   export const on types, two-argument @ptrCast(*T, x), and four-argument
   expectApproxEqAbs(f64, …). None of these have existed in the language
   since roughly 0.11.

The part that matters

After all of that, zig build test passed — running zero tests:

$ zig test src/root.zig
All 0 tests passed.

Every domain root says pub const formulas = @import("formulas.zig"); and
nothing references it. Zig analyses a top-level declaration only when
something does, so those files were never part of the compilation and their
test blocks did not exist. 640 of them.

Adding one reference per import — test { _ = formulas; } — changes the
measurement:

before:  All 0 tests passed.
after:   253 passed; 1 skipped; 0 failed.   (254 reachable)

The one failure

test "Barbero-Immirzi prediction" failed the moment it could run.
quantumProjection can only ever return 0.436992, 0.618034 or 1.236068,
because |c[4]| + |c[5]| is 0, 1 or 2 for every E8 root — while the physical
value is 0.2375. The only value inside the asserted range is the fallback
branch taken when the projection has no input.

Filed as #3 and marked skipped with a pointer, not deleted and not relaxed:
the assertion is the correct physics and the projection is what has to
change. Fixing it means choosing a mapping, and any mapping I chose would be
fitted to the assertion it had to satisfy.

Not fixed here

src/plasma/ has no formulas.zig (#2) — the only domain without one. Its
root.zig exports nothing but testValues, which imports the missing file,
so the module has never had an implementation. The re-export is commented out
with the reason; writing the formulas from test_values.zig would produce a
model that agrees with its own test by construction.

CI

Three steps: zig build, zig build test, and a guard that fails when fewer
than 200 tests run. The third exists because this repository proves that a
suite running nothing also exits 0.