feat(workload): define canonical workload wire format

[zzylol]

Why

ASAPPlanner and downstream components need one wire representation for the existing canonical QueryWorkload and DataWorkload types. Copying them into a second DTO would let recurrence, time selection, accuracy, evidence freshness, and data-arrival semantics drift.

What

• Adds Serde contracts directly to the canonical workload types; it does not introduce a separate snapshot envelope or duplicate workload model.
• Keeps scalar newtypes such as timestamps, durations, rates, confidence, queries, and repetition intervals transparent.
• Gives acronym-bearing languages and SQL dialects explicit stable names: promql, sql, datafusion, elastic_dsl, datafusion_sql, clickhouse_sql, and elastic_sql.
• Rejects unknown fields on workload structs and struct-style enum variants, including nested AccuracyTarget::EpsilonDelta.
• Adds representative strict JSON round-trip, exact wire-name, and schema-drift regression tests. The Serde contract is also usable by YAML transports.

Boundary

Deserialization establishes only the wire shape. Callers must still invoke QueryWorkload::validate before planning; semantic validation rejects invalid rates, recurrence, confidence, latency, and data-arrival combinations. Evidence freshness is intentionally time-dependent and is checked separately through Evidence::is_fresh_at(planning_time), not during deserialization or static workload validation.

This PR intentionally serializes the existing source-of-truth workload enums instead of defining new streaming or deployment enums.

Verification

• cargo test -p asap-types workload::tests
• cargo clippy -p asap-types --all-targets -- -D warnings
• cargo fmt --all -- --check

This PR is independently based on #338; it is not a prerequisite of #339.