Import AbstractMCMC to fix package precompilation (UndefVarError)

[ChrisRackauckas-Claude]

Problem

main CI is red across four checks (Core, Datafit, QA ×2 Julia 1/lts, and Documentation build). All fail at the same root cause: the package no longer precompiles.

ERROR: LoadError: UndefVarError: `AbstractMCMC` not defined in `EasyModelAnalysis`
Hint: AbstractMCMC is loaded but not imported in the active module Main.
  @ ~/.../src/datafit.jl:262

The mcmcensemble keyword of both bayesian_datafit methods is annotated AbstractMCMC.AbstractMCMCEnsemble (added in commit 6a350b8, July 2023), but AbstractMCMC was never imported into the module. It was only reachable transitively because older Turing versions re-exported the bare AbstractMCMC binding. The Turing version now resolved in CI no longer brings AbstractMCMC into the module's scope, so the type annotation fails at precompile/load time and every test group + the docs build dies before running.

Fix

Declare AbstractMCMC as a direct dependency and using AbstractMCMC. This is the symbol the source actually references. AbstractMCMC was already present transitively in the resolved environment (v5.15.1), so no manifest/version churn beyond declaring it. Compat set to 4, 5 (the AbstractMCMCEnsemble abstract type has been stable across both).

Local verification (Julia 1.10)

Reproduced the failure on a clean main checkout (package failed to load), then confirmed the fix:

  23504.7 ms  ✓ EasyModelAnalysis
RESULT: LOADED OK
AbstractMCMCEnsemble resolves: AbstractMCMC.AbstractMCMCEnsemble
bayesian_datafit methods: 2

The package now precompiles and loads, and both bayesian_datafit method signatures resolve.

Please ignore until reviewed by @ChrisRackauckas.