Add LogUniform distribution to beast.base.spec

[alexeid]

Summary

Adds LogUniform as a proper, scale-invariant ScalarDistribution for strictly positive quantities — the defensible replacement for the improper OneOnX prior, which was deliberately not migrated to the spec types (see existing note in version.xml:226).

• Density p(x) = 1 / (x * log(upper/lower)) on [lower, upper], equivalent to Uniform on log(x).
• Apache Commons Statistics has no reciprocal distribution, so ContinuousDistribution is implemented inline with closed-form density, CDF, inverse CDF, mean, variance, and inverse-CDF sampler.
• Parameters lower, upper are RealScalar<PositiveReal>; the distribution validates 0 < lower < upper < infinity in refresh().
• Registered as a BEASTInterface service provider in both module-info.java and version.xml.

Motivation

FLC's beast3 migration (4ment/flc#10) uses OneOnX on the coalescent popSize; that block is currently commented out because OneOnX has no spec-type counterpart. LogUniform on e.g. [1e-6, 1e6] gives twelve orders of magnitude of scale-invariant prior support — behaviourally equivalent for typical molecular-clock analyses, philosophically cleaner (proper, integrable, has a CDF and quantile), and usable anywhere ScalarDistribution is expected.

Test plan

• LogUniformTest (8 tests): closed-form density, zero outside support, CDF/quantile round-trip (1000 random pairs), log-space median equals sqrt(ab), mean formula, calculateLogP via param, XML parse, rejection of degenerate bounds
• Full beast-base suite: 399 tests, 0 failures, 0 errors
• Verify downstream: use LogUniform for popSize prior in FLC example XML