feat(security): per-tenant isolation for spend, secrets, strict mode

[Destynova2]

Summary

Closes 3 multi-tenant isolation gaps flagged by PR #326's regression suite. The 3 `#[ignore]`'d tests now pass.

• Per-tenant budget: `SpendTracker` grows `check_tenant_budget(tenant, …)` backed by per-tenant in-memory caches and per-tenant JSONL journals at `spend//.jsonl`. `record_tenant` no longer also accumulates into the global counter, so a tenant overspend cannot trip the global budget for other tenants.
• Per-tenant secrets: `SecretBackend::get` takes `(tenant, name)`. All three backends (LocalEncrypted, Env, File) resolve `/` first and fall back to the legacy flat layout for global secrets so single-tenant deployments need no migration.
• `strict_tenant` flag: new `[security] strict_tenant` config field plus `tenant_required_middleware` that returns HTTP 400 with a `missing_tenant` body when the flag is on and no tenant id can be resolved (X-Tenant-ID header, JWT claim, or virtual key).
• Tenant id source priority: VirtualKeyContext > JWT claim > X-Tenant-ID header. Header is only consulted when no authenticated tenant exists, so a client cannot impersonate a JWT tenant.

API breaking changes (all internal except `SecretBackend` trait)

• `SecretBackend::get`: now takes `&str tenant` before `&str name`.
• `SpendTracker` grows `check_tenant_budget` plus `DEFAULT_TENANT` bucket.
• `SpendTracker::record_tenant` stops also calling `self.record(...)`.

Test plan

• `cargo test --lib --tests` — 266 passed, 1 ignored (anomaly detector waiting on feat(security): tool-call spike anomaly detector (T-AD1) #308)
• `cargo test --test lib multi_tenant` — 8 multi-tenant integration tests pass; previously 3 were `#[ignore]`d as known gaps
• `cargo clippy --lib --tests -- -D warnings` — clean
• Single-tenant deployments unaffected (legacy flat layout fallback verified)

Origin

This commit was authored by background agent `a6ab9ef6391d6f6f5` on 2026-04-28 but the agent hit org usage cap before opening the PR. Commit content unchanged from `2fe3c1c`; cherry-picked onto current main (which already includes #326's regression suite).

🤖 Generated with Claude Code

[github-actions]

Mutation testing (PR diff sample)

Informational — never blocks merge. Full matrix runs on main.

Metric	Value
Status	missed
Duration	757 s
Total	14
Caught	0
Missed	0
Timeout	0
Unviable	0

Legend: clean (no survivors), missed (inspect artifact), timed-out (25 min cap reached).

Artifact: mutants-pr-results-52b2664853a86d444469cf3f504709a97c88c701.