feat: implement CMS Type-1 MET correction in CorrectedMETFactory

[alefisico]

Implements the full CMS Type-1 MET correction recipe in CorrectedMETFactory, addressing the remaining open item from #1064. (This feature was asked by @mcremone)

The previous fix (commit 6dfd4ad) corrected the sign and _orig reference bugs, but the factory still did not implement the proper Type-1 correction: it was missing muon subtraction, L1-only JEC, EM fraction cuts, and the CorrT1METJet low-pT jet collection.

This PR adds a new Type-1 mode that follows the official CMS procedure as described in the JME General Meeting slides (Feb 2026):

1. Compute muon-subtracted raw pT and phi for each jet
2. Apply L1-only and L1L2L3 JEC to the muon-subtracted raw pT
3. Select jets passing pt_noMuL1L2L3 > 15 GeV and EM fraction < 0.9
4. Vectorial correction: RawMET -= Σ(pt_noMuL1L2L3 - pt_noMuL1) for both Jet and CorrT1METJet collections

Behaviour

Type-1 mode is opt-in and backward compatible. Legacy behavior is completely unchanged when the new parameters are not provided.

# Construct with L1-only and L1L2L3 JEC correctors
met_factory = CorrectedMETFactory(name_map, jec_L1L2L3=jec_L1L2L3, jec_L1=jec_L1)

# Build with RawMET and (optionally) CorrT1METJet
corrected_met = met_factory.build(
    events.MET, corrected_jets,
    in_RawMET=events.RawMET,
    in_CorrT1METJets=events.CorrT1METJet,
)

The JEC correctors are duck-typed — both FactorizedJetCorrector and CorrectionLibJEC work.

Implementation details

• JEC factors are computed once and reused across all JES/JER systematic variations
• Helper functions return awkward.zip records (not tuples) for dask-awkward compatibility
• Arrays are flattened before JEC calls and unflattened after, for CorrectionLibJEC compatibility
• JES/JER variations use the ratio varied_pt / nominal_pt to scale pt_noMuL1L2L3, and re-apply the > 15 GeV cut
• CorrT1METJet contribution stays nominal for JES/JER variations (it's not in corrected_jets)