fix: deflake //rs/dogecoin/ckdoge/minter:integration_tests

[mraszyk]

This PR attempts to deflake the test //rs/dogecoin/ckdoge/minter:integration_tests.

PLEASE REVIEW EXTRA CAREFULLY since this PR has been fully Claude-generated.

Now I understand the root cause completely:

1. last_median_fee_per_vbyte is initialized to FeeRate::from_millis_per_byte(1) in the minter state
2. It only updates when estimate_fee_per_vbyte successfully calls the Dogecoin canister
3. During setup, RefreshFeePercentiles fires while the Dogecoin canister is still syncing blocks → fails → last_median_fee_per_vbyte stays at 1
4. estimate_withdrawal_fee (called before the transaction is built) uses the stale 1 value
5. Later, ProcessLogic fires after the Dogecoin canister is synced → succeeds → updates last_median_fee_per_vbyte to DEFAULT_REGTEST_FEE → transaction built with dogecoin_fee:
   227000000

The fix: use fee_estimator.estimate_median_fee(&s.last_fee_per_vbyte) instead of s.last_median_fee_per_vbyte. For Regtest, estimate_nth_fee ignores fee percentiles and always returns
DEFAULT_REGTEST_FEE, so this correctly uses DEFAULT_REGTEST_FEE even before the first successful fee refresh.