feat: add round-relative oracle timestamp window

[owohdora-stack]

Summary

Closes #265
Replace the absolute 300-second oracle data freshness check with a round-relative timestamp window that requires oracle payload timestamps to fall within the round's economic activity period.

Why

The old absolute-freshness check (now - timestamp ≤ 300s) could accept wrong-phase prices — a price from well after the round ended could still be "fresh" relative to the current ledger time. The new window anchors around the round's start and estimated end:

[lower, upper] = [round_start - skew, round_end_estimate + skew]

where round_end_estimate = round_start + (end_ledger - start_ledger) × 5s.

What changed

File	Change
types.rs	Added start_timestamp: u64 to Round; added OracleTimestampSkew to ConfigChangeKind/ConfigChangePayload
betting.rs	Record env.ledger().timestamp() when creating a round
common.rs	Added SECONDS_PER_LEDGER, DEFAULT_ORACLE_TIMESTAMP_SKEW, MIN/MAX_ORACLE_TIMESTAMP_SKEW
errors.rs	Added OracleTimestampOutsideWindow = 66
settlement.rs	Replaced [future + stale] checks with [future + round-window] validation using configurable skew
config.rs	Added schedule_oracle_timestamp_skew, get_oracle_timestamp_skew, validation, and _apply_config_payload/_current_config_payload handling
contract.rs	Exposed schedule_oracle_timestamp_skew and get_oracle_timestamp_skew publicly
security.rs	Updated existing tests; added 4 new boundary tests (before window, after window, lower edge, upper edge)
event_coverage.rs	Updated stale-data test to expect OracleTimestampOutsideWindow
ORACLE_OPERATOR_RUNBOOK.md	Updated validation order, timestamp field docs, troubleshooting matrix, config table, and Playbook B

Acceptance criteria

• Pre-bet timestamp rejected (test_resolve_round_timestamp_before_round_window)
• In-window timestamp accepted (test_resolve_round_valid_payload, boundary tests)
• After-round timestamp rejected (test_resolve_round_timestamp_after_round_window)
• Future timestamps still rejected (test_resolve_round_future_timestamp)
• Configurable skew via timelock (schedule_oracle_timestamp_skew)
• Docs updated

Testing

The codebase has a pre-existing LengthExceedsMax issue in the DataKey enum (51 variants exceed Soroban's contracttype limit) which prevents compilation of both main and this branch. The changes are structurally sound — all new code avoids adding DataKey variants by using instance storage with Symbol keys.

[josephchimebuka]

@owohdora-stack Pls can you fix Conflicts and send a message in the group saying you've fixed them so I can merge