Dependency-boundary shim library (ObjexxFCL Array, Constant::*, libm intrinsics)

[Tokarzewski]

Part of #64.

Problem

Most compile-closure failures are not the function itself — they are unported types/symbols it references. Porting each fully is wasteful; a thin shim satisfies the compiler and is written once, reused everywhere.

Shims to build/extend

• ObjexxFCL Array1D/Array2D → List-backed 1-based arrays in ep_prelude.mojo, mirrored by std::vector-backed structs in ep_oracle.h (C++ keeps a(i)/a(i,j) call syntax, Mojo uses []; both 1-based). ✅ landed main 61db4bf
• Constant::* → extended (added Sigma, UniversalGasConstant, Gravity); kept Mojo/C++ values identical. ✅
• libm → verified empirically: sin links under -Xlinker -ldl (no -lm) — trig links as intrinsic (myth busted; matches the libm-link finding). ✅
• Psychrometric / table lookups behind a shim interface. ⏳ only the pvstar stub exists; real psychro shims still TODO — keep open.

Done when

• A shimmed dependency compiles a dependent fn without a full port.
• The shim layer in scripts/sft/diff_verify_ep.py consumes these uniformly (and its dead energyplus-mojo/.pixi Mojo path is fixed → MOJO_HOME-overridable, live NuMojo default).
• Each shim verified vs C++ once: an Array1D + Constant.Sigma + sin function compiles + numerically matches C++ through diff_verify_ep.py (1/1). Existing items unaffected (9/10 spot-check; the 1 fail is a pre-existing std::cout-without-<iostream> harness gap).

Remaining: psychrometric shims. Otherwise the dependency-boundary shim library is in place.

🤖 Generated with Claude Code

[Tokarzewski]

Progress — first real psychrometric shim landed (PR #75, draft).

The last open box (psychrometric / table lookups behind a shim interface) is now
started: PsyPsatFnTemp (water saturation pressure) is a real shim instead
of the pvstar constant stub, ported faithfully from the oracle's
PsyPsatFnTemp_raw (Psychrometrics.cc, non-IF97 branch). Python reference +
Mojo/C++ shim codegen + 12 pytest cases verified against textbook saturation
pressures (611 Pa @ 0 °C, ~101325 Pa @ 100 °C, ASHRAE 3.17 kPa @ 25 °C).

Toolchain-gated remainder (deferred — env here has no Mojo + g++):

1. Final shim gate. Run the emitted Mojo + C++ shims through
   scripts/sft/diff_verify_ep.py exactly as the Array1D + Constant.Sigma + sin
   shim was verified (compile both standalone, sample inputs across
   [-100, 200] °C incl. the branch boundaries 173.15 K / triple point / 473.15 K,
   assert numeric match). Wire PsyPsatFnTemp into the prelude shim set so a
   dependent function (e.g. PsyRhovFnTdbRh) compiles against it.
2. Remaining psychro shims, same Python-reference→codegen→diff_verify pattern,
   in dependency order:
   • PsyTsatFnHPb (sat temp from enthalpy+pressure — iterative; the _raw)
   • PsyWFnTdbTwbPb (humidity ratio from dry/wet bulb + pressure)
   • PsyRhoAirFnPbTdbW, PsyHFnTdbW, PsyTwbFnTdbWPb
     Each is a self-contained scalar fn in Psychrometrics.cc; the cache/error
     wrappers are not needed for a shim.
3. Replace the pvstar stub call sites once PsyPsatFnTemp is gated green.

The Python math + codegen are the pytest-verifiable slice done now; items 1–3
are pure build/verify chores once a Mojo+g++ box is available.