feat: add log2 log10 sinh cosh tanh asinh acosh atanh to the formula language

[hozblok]

Closes TODO src/cpp/csconstants.hpp:222 (// TODO add log2 log10 sinh cosh tanh asinh acosh atanh).

Scope

Eight new one-arg math functions exposed in the formula language, both on the real and complex evaluation paths, with chain-rule derivatives where applicable.

function	derivative	divergence note
log2(x)	1 / (x · ln 2)	throws on x = 0
log10(x)	1 / (x · ln 10)	throws on x = 0
sinh(x)	cosh(x)	—
cosh(x)	sinh(x)	—
tanh(x)	1 / cosh(x)²	—
asinh(x)	1 / √(1 + x²)	—
acosh(x)	1 / √(x² − 1)	throws on `
atanh(x)	1 / (1 − x²)	throws on `

Wiring

• csconstants.hpp — using boost::multiprecision::{log2, log10, sinh, cosh, tanh, asinh, acosh, atanh};
• cseval.hpp / cseval_complex.hpp — static _X (value) + static _X_d (derivative) per function.
• cseval.cpp / cseval_complex.cpp — new entries in functionsOneArg and functionsTwoArgsDLeft for each.

Complex log2 / log10

Boost's cpp_complex_backend does not provide native log2 / log10. Implemented manually via the change-of-base identity log_b(z) = ln(z) / ln(b), using the existing _log primitive. Derivative uses the same identity for the constant denominator. Real log2 / log10 go through Boost's native overloads on cpp_dec_float.

Divergences

acosh / atanh derivatives are singular at x = ±1. They throw std::invalid_argument (Python-side ValueError) with the function name in the message, matching the existing _asin_d / _tan_d / _log_d convention.

Tests

Eight new files under tests/functions/one_arg/, one per function (test_log2.py … test_atanh.py). Each covers:

• value at a convenient input (e.g. log2(8) → 3, sinh(0) → 0)
• derivative at a convenient input where defined
• divergence on |x| = 1 for acosh/atanh

Matches the existing terse test_X.py shape (without the long precision-256 golden values — those are easy to add later if desired).

Full suite 410/410 (+22 new), 3 xfailed unchanged.