feat(v2): implement TTL policy for all v2 record types

[collinsezedike]

Closes #72

Summary

Implements TTL policy for all four of v2's long-lived record types (AssertionV2, Resolution, Position, Credit), per #72's scope.

Every write already re-extended TTL via a flat 30-day bump (INSTANCE_BUMP_AMOUNT), matching v1's constant. That's a hardcoded margin, not one sized from the deployment's actual policy: anti_snipe_hard_max_secs isn't capped today (that's #114/#116, still in review), and even once bounded at 29 days, anti_snipe_hard_max_secs + reveal_duration_secs can exceed 30 days on its own for a deployment configured with a long anti-snipe window and a long reveal window. A record whose TTL is sized independently of the round it belongs to can expire mid-round.

What changed

• New record_bump(policy) helper computes the TTL bump (threshold, amount) in ledgers from that specific assertion's own pinned PolicySnapshotV2: anti_snipe_hard_max_secs + reveal_duration_secs + SETTLEMENT_GRACE_SECS (a fixed 7-day settlement/withdrawal grace period), floored at INSTANCE_BUMP_AMOUNT so this never shrinks the existing margin, only ever matches or exceeds it.
• set_assertion, set_resolution, set_position, and add_credit all now compute their bump from this instead of the flat constant.
• New permissionless bump_ttl(id, address). Position/Credit keys aren't enumerable on-chain (the design deliberately avoids an unbounded voter vector), so a record nobody happens to touch again before its own next write, a registered voter who never reveals, or unclaimed credit sitting past settlement, previously only got its TTL renewed if that address's own owner called reveal/settle/withdraw. Anyone who knows the (id, address) key can now renew it directly. A no-op (not an error) for any of the four entries that don't exist for a given id/address; fails only with AssertionNotFound if id itself doesn't exist.

On the issue's other requirements

• Restoration path: Soroban's archival-restoration mechanic (a client-submitted RestoreFootprintOp) is transaction-level, not something contract code can trigger internally. Nothing in settle/withdraw/get_* assumes non-restorability, so a restored entry works normally once the client includes it in a transaction's footprint. bump_ttl reduces how often restoration is even needed by making proactive renewal permissionless.
• Indexable events for key discovery: checked against [Feature] V2: implement reveal phase and commitment verification #67-[Feature] V2: implement credit accrual and owner-authorized withdrawal #70's events rather than assumed. Every Position/Credit key is already derivable without on-chain enumeration: fixed positions' keys come from Asserted+Disputed (both carry the address), external positions' from PositionFunded, and credit's from Settled/DustCredited (both carry the address). No new event needed for this requirement.

Test plan

• cargo fmt --check, cargo clippy --workspace --all-targets -- -D warnings, and cargo test --workspace pass locally (104 tests in tholos-v2, all 100 pre-existing ones unchanged and passing, confirming the floor keeps default-policy behavior identical to before)
• CONTRACT_V2.md updated: the TTL section now describes the sized calculation, and bump_ttl has its own function reference entry
• Not applicable: scripts/testnet-smoke.sh covers v1 only
• New tests: test_dispute_sizes_resolution_and_position_ttl_from_policy_when_larger_than_instance_bump (confirms a large anti_snipe_hard_max_secs produces a TTL bump exceeding the flat floor, with the exact expected ledger count), test_bump_ttl_extends_assertion_resolution_position_and_credit (full flow to all four record types populated, confirms bump_ttl restores every one to a full bump after partial decay), test_bump_ttl_on_nonexistent_assertion_fails, test_bump_ttl_is_noop_when_position_and_credit_do_not_exist_for_address