fm: extract shared keyed-case adoption scaffolding from the saga and certificate engines

[smklein]

The saga and certificate diagnosis engines (nexus/fm/src/diagnosis/saga.rs and certificate.rs) each carry a near line-for-line copy of the same bookkeeping around parent-forwarded cases:

• an UninterpretableCase enum (ForeignFact, Disagreeing<Key>s, NoFacts)
• the loop that closes uninterpretable cases with a "cannot interpret case" comment and a "closing uninterpretable case" warning
• the BTreeMap Entry::Vacant/Entry::Occupied loop that dedups cases by key, keeping the lowest CaseUuid and closing the rest as duplicates

The uninterpretable half also appears a third time in physical_disk.rs. Per-engine fact modeling (what a parsed case's payload looks like) is legitimately per-engine under RFD 603, but this adoption scaffolding is engine-agnostic, and a fix to the tiebreak or the close messages today has to be made in three places.

Proposal: a helper in diagnosis/mod.rs along the lines of

fn adopt_keyed_cases<K: Ord + Display, P>(
    builder: &mut SitrepBuilder<'_>,
    de: DiagnosisEngineKind,
    parse: impl Fn(&Case) -> Result<(K, P), impl Display>,
) -> BTreeMap<K, (CaseUuid, P)>

that saga.rs and certificate.rs call, keeping only their own parse_case bodies. physical_disk.rs could adopt the uninterpretable half in the same change or a follow-up. Existing tests in all three engines already cover uninterpretable and duplicate cases.

Deferred from review of #11238 to keep that PR scoped to the certificate engine.

[smklein]

dup of #11246