Add BioPreDyn-bench B1, B2, B4, B5 benchmarks

[ChrisRackauckas-Claude]

Summary

• Adds 4 of 6 BioPreDyn-bench work-precision benchmarks for systems biology ODE solver comparison
• B1: S. cerevisiae genome-wide kinetics (285 ODEs, 1759 params) - uses SBMLImporter + JLD2 parameter file
• B2: E. coli central carbon metabolism (18 ODEs, 116 params) - uses SBMLImporter + JLD2 parameter file
• B4: CHO cell metabolism (35 ODEs, 117 params) - manually translated from Matlab (SBML uses rate rules)
• B5: EGF/TNFa signal transduction logic network (26 ODEs, 124 params) - manually translated from C source
• Updates Bio/Project.toml: adds SBMLImporter dependency, widens Catalyst/MTK compat bounds
• B3 (E. coli gene regulation, 47 ODEs with diauxic shift events) deferred due to SBMLImporter callback _to_float bug and parameter mapping issues
• B6 (spatial PDE) skipped as not suitable for standard ODE benchmarks

Each benchmark includes: model loading, solution visualization, reference solution generation, and work-precision diagrams at high/low tolerances with SDIRK methods and a summary comparison.

Closes #555

Test plan

• B1 solves with CVODE_BDF at 1e-5 and 1e-12 tolerance
• B2 solves with CVODE_BDF, reference solution works, WorkPrecisionSet verified
• B4 solves with Rodas5P at 1e-8 and 1e-14 tolerance
• B5 solves with Rodas5P at 1e-8 and 1e-14 tolerance
• CI weave of all 4 benchmarks

🤖 Generated with Claude Code

[ChrisRackauckas-Claude]

Build Artifact Review

Reviewed the CI build artifacts for B4 and B5 benchmarks:

B4 (BioPreDynB4.jmd) - All Good

• Solution plot renders (35 ODEs, one dominant variable at ~10^5 scale)
• All 4 work-precision diagrams (High Tol, Low Tol, SDIRK, Summary) render correctly
• No errors

B5 (BioPreDynB5.jmd) - Fixed one issue

• Solution plot had DimensionMismatch error: Plots.jl's ODE recipe tried to arrange 26 variables into subplot grid with uneven rows (10, 10, 6), causing DimensionMismatch: number of columns of each array must match (got (10, 10, 6)). Fixed by plotting Array(sol)' directly against sol.t instead of using the ODE plot recipe.
• All 4 work-precision diagrams rendered correctly with no issues
• Reference solution generates fine (0.3s)

Julia Tests CI failure - Not related to this PR

The "Julia Tests" check failed with No space left on device during Julia setup - infrastructure issue on the CI runner, not related to PR changes.

Fix pushed

Commit 453933d fixes the B5 solution plot error.