[BugFix][Target][LLVM] Route sinh/cosh/atan/asinh/erf through libm extern

[swjng]

Summary

Six LLVM legalize rules in src/target/llvm/intrin_rule_llvm.cc use inline mathematical identities that fail on representable inputs because the intermediate computation overflows or cancels, even though the true result is in float32 range:

Op	Inline form	Failure	True result
sinh/cosh (#19559)	(exp(x) ± exp(-x)) / 2	exp(89) > FLT_MAX, intermediate is inf	sinh(89) ≈ 2.24e38
atan (#19560)	asin(x / sqrt(x²+1))	x² overflows for `	x
asinh (#19561)	log(x + sqrt(x²+1))	same x² overflow → log(inf)=inf	asinh(3e22) ≈ 52.45
erf (#19562)	A&S 1 − poly(t)·exp(−x²)	poly·exp(−x²) ≈ 1 for tiny `	x
acosh (no issue)	log(x + sqrt(x²−1))	same x² overflow → inf	acosh(3e22) ≈ 52.45

acosh was not in the original issue cluster but shows the identical bug pattern to asinh; folding it in keeps this PR's scope consistent ("naive math identity → libm extern"). Happy to split it out if reviewers prefer.

Fix

Route all six through the existing DispatchPureExtern<FloatSuffix> helper — i.e. sinhf, coshf, atanf, asinhf, acoshf, erff — the same pattern asin/acos use after #19567. ULP-grade accuracy across the reported ranges.

sinh(89.0):    ORT=2.244806e+38  TVM=2.244806e+38  (was inf)
atan(3e22):    ORT=1.5707964     TVM=1.5707963     (was 0.0)
asinh(3e22):   ORT=52.44863      TVM=52.44863      (was inf)
acosh(3e22):   ORT=52.44863      TVM=52.44863      (was inf)
erf(3e-12):    ORT=3.385e-12     TVM=3.385e-12     (was 0.0)

Atan is re-enabled in test_unary; the overflow that previously broke it is fixed.

Notes for reviewers

Inline-vs-extern decision. If the inline identities were a deliberate fast-path (e.g. for autovectorization or to avoid extern-call overhead in tight loops), please flag it and I'll switch to stable inline forms instead — exp(x − ln 2) ± exp(−x − ln 2) for sinh/cosh, range-reduced asinh/acosh sign(x)·log(2|x|) for large |x|, small-|x| Taylor branch for erf, etc. I could not find evidence of such intent in the git history (sinh/cosh: original commit; atan/asinh/acosh: #17945 / #17969 follow-ups; erf: #18104 was framed as "more precise than tanh-approx", not "fast inline").

Fixes #19559.
Fixes #19560.
Fixes #19561.
Fixes #19562.

[gemini-code-assist]

Code Review

This pull request refactors the LLVM legalization rules for several intrinsic operations, including cosh, sinh, atan, asinh, and erf, by replacing manual mathematical expansions with a unified DispatchPureExtern call. Additionally, the Atan operator has been enabled in the ONNX frontend tests. The reviewer pointed out that the using namespace intrin; statements, local call variable assignments, and manual null checks are redundant in the new implementation because DispatchPureExtern is fully qualified and performs its own internal validation.