perf(diagnostics): bound RPC latency aggregation

[beruro]

Problem

RPC latency aggregation and diagnostics retention can grow cardinality or preserve more raw detail than needed, increasing memory/persistence cost and privacy risk.

Solution

• Bound RPC duration aggregation cardinality while preserving stable buckets.
• Sanitize runtime counters before persistence/submission.
• Retain only diagnostics detail required for performance investigation.
• Add bucket-edge, queue, and sanitization coverage.
• Validation: focused Vitest, TypeScript typecheck/ESLint, Rust usage-diagnostics tests/check/rustfmt, and diff check.

Potential risks

• Over-aggressive sanitization or bucketing could remove useful diagnostic distinctions.
• Bucket boundaries must remain stable for comparisons across versions.
• This PR is independent of the async-resource and session stacks.

Verification

Previous GitHub CI on head ebd23423d:

• pnpm install --frozen-lockfile — PASS in Frontend (typecheck · lint · test).
• pnpm typecheck — PASS in Frontend (typecheck · lint · test).
• pnpm lint — PASS in Frontend (typecheck · lint · test).
• pnpm run test — PASS in Frontend (typecheck · lint · test).
• node scripts/tauri/prepare-sidecars.cjs --profile debug — PASS in Rust (clippy).
• (cd src-tauri && cargo clippy --workspace --all-targets -- -D warnings) — PASS in Rust (clippy).
• AI-attribution history scan — PASS.

Current GitHub CI is re-running against the latest develop merge for head 173101a6a8; those results remain required before merge.