perf: use String.compare in ConfigEval decision tree

[kmill]

This PR improves performance of matching configuration items by using String.compare to implement the decision tree for interpreting configuration option names.

#13796 optimized String.compare to do only a single memcmp, which we can take advantage of here.

[mathlib-lean-pr-testing]

Mathlib CI status (docs):

• ❗ Batteries/Mathlib CI will not be attempted unless your PR branches off the nightly-with-mathlib branch. Try git rebase ada1696f040c749e6038e516266c4c1c94eb27e1 --onto 43ef70db63e2c1c1720b84fd153125c96a00c5d2. You can force Mathlib CI using the force-mathlib-ci label. (2026-05-20 17:57:27)
• ✅ Mathlib branch lean-pr-testing-13806 has successfully built against this PR. (2026-08-29 19:04:19) View Log

[leanprover-bot]

Reference manual CI status:

• ❗ Reference manual CI will not be attempted unless your PR branches off the nightly-with-manual branch. Try git rebase ada1696f040c749e6038e516266c4c1c94eb27e1 --onto 803553a556fd82fa1060efb0c43eda542130cb16. You can force reference manual CI using the force-manual-ci label. (2026-05-20 17:57:29)
• ✅ Reference manual branch lean-pr-testing-13806 has successfully built against this PR. (2026-08-29 18:14:26) View Log
• 🟡 Reference manual branch lean-pr-testing-13806 build against this PR didn't complete normally. (2026-08-29 18:15:59) View Log

[kmill]

!bench

[leanprover-radar]

Benchmark results for 6a68eea against ada1696 are in. No significant results found. @kmill

• 🟥 build//instructions: +1.7G (+0.02%)

Small changes (10🟥)

• 🟥 build/module/Lean.Elab.ConfigEval.MetaInstances//instructions: +79.1M (+3.33%) (reduced significance based on absolute threshold)
• 🟥 build/module/Lean.Elab.ConfigEval.Util//instructions: +131.3M (+3.44%) (reduced significance based on *//lines)
• 🟥 build/module/Lean.Elab.Tactic.BVDecide.Frontend.Attr//instructions: +73.7M (+2.06%) (reduced significance based on absolute threshold)
• 🟥 build/module/Lean.Elab.Tactic.Do.VCGen.Basic//instructions: +77.2M (+0.85%)
• 🟥 build/module/Lean.Elab.Tactic.Grind.Config//instructions: +114.8M (+1.33%) (reduced significance based on absolute threshold)
• 🟥 build/module/Lean.Elab.Tactic.Lets//instructions: +53.0M (+1.31%)
• 🟥 build/module/Lean.Elab.Tactic.LibrarySearch//instructions: +23.8M (+0.64%)
• 🟥 build/module/Lean.Elab.Tactic.NormCast//instructions: +200.7M (+1.66%) (reduced significance based on absolute threshold)
• 🟥 build/module/Lean.Elab.Tactic.Simp//instructions: +167.9M (+0.87%) (reduced significance based on absolute threshold)
• 🟥 build/module/Lean.Elab.Tactic.SolveByElim//instructions: +65.7M (+1.45%) (reduced significance based on absolute threshold)

[kmill]

!bench

[leanprover-radar]

Benchmark results for e604bbb against e991a05 are in. No significant results found. @kmill

• 🟥 build//instructions: +2.6G (+0.02%)

Small changes (9🟥)

• 🟥 build/module/Lean.Elab.ConfigEval.MetaInstances//instructions: +98.7M (+3.79%) (reduced significance based on absolute threshold)
• 🟥 build/module/Lean.Elab.ConfigEval.Util//instructions: +141.8M (+3.61%) (reduced significance based on *//lines)
• 🟥 build/module/Lean.Elab.Tactic.Do.VCGen.Basic//instructions: +81.4M (+0.90%)
• 🟥 build/module/Lean.Elab.Tactic.Grind.Config//instructions: +142.6M (+1.59%) (reduced significance based on absolute threshold)
• 🟥 build/module/Lean.Elab.Tactic.Lets//instructions: +57.1M (+1.37%)
• 🟥 build/module/Lean.Elab.Tactic.NormCast//instructions: +249.9M (+2.16%) (reduced significance based on absolute threshold)
• 🟥 build/module/Lean.Elab.Tactic.Simp//instructions: +201.6M (+1.09%)
• 🟥 build/module/Lean.Elab.Tactic.SolveByElim//instructions: +70.5M (+1.52%)
• 🟥 build/module/Lean.Meta.Tactic.BVDecide.Attr//instructions: +61.8M (+1.69%)