fix(l1): bump EIP-1559 fees on same-nonce poster retries

[stephenctw]

Summary

Fixes same-nonce batch-poster retries that could rebroadcast underpriced EIP-1559 replacements when the fee market was flat. The poster now tracks in-flight fees per wallet nonce and floors each re-estimate against a shared ≥10% replacement bump (max fee ×1.1 +1; priority ×2), matching the mempool flusher’s rule.

• Hoist bumped_replacement_fees into l1::eip1559 and add fees_for_nonce so poster + flusher share one replacement policy.
• On successful broadcast, record fees; on failed send, leave the floor unchanged; prune entries once Latest advances past the nonce.
• Cover bump math, mixed-component floors, Latest prune, and failed-send non-record behavior with unit/Anvil tests.

Test plan

• cargo test -p sequencer --lib eip1559
• cargo test -p sequencer --lib submitter::poster
• Confirm flusher still compiles/uses shared bumped_replacement_fees (cargo test -p sequencer --lib recovery::flusher)
• Anvil-backed poster replacement test (submit_batches_replacement_clears_ten_percent_bump)

[stephenctw]

@jplgarcia FYI

[stephenctw]

Fixes the permanent poster wedge from asymmetric replacement bumps; limits re-escalation to the head nonce; adds geth underpriced fast-path floor raise.

[jplgarcia]

Thanks for the fix @stephenctw . I validated this against the current PR head with a real pending transaction on Anvil (--no-mining).

The fee-bump logic itself is correct: the retry uses the same nonce and raises both EIP-1559 values by at least 10%:

• maxFeePerGas: 2000000001 -> 2200000002
• maxPriorityFeePerGas: 1 -> 2

However, the poster still does not reach eth_sendRawTransaction in the real replacement path.

The failure occurs earlier, during gas estimation:

eth_estimateGas(
  from=<poster>,
  nonce=0,
  maxFeePerGas=2200000002,
  maxPriorityFeePerGas=2,
  block="pending"
) -> -32003 "nonce too low"

At that point, nonce 0 is already pending (latest=0, pending=1). Anvil applies its mempool nonce policy to the pending gas simulation and rejects it, so the replacement is never signed or broadcast.

As a control, I estimated the same addInput call without an explicit nonce, fixed that gas value on the replacement transaction, and submitted it with nonce 0 and the bumped fees. Anvil accepted the replacement, replaced the pending transaction, and it mined once blocks were resumed.

I think the remaining change should be scoped to the same-nonce replacement path:

1. Estimate gas without the pending nonce.
2. Set the resulting gas explicitly on the final replacement transaction.
3. Keep the current per-nonce fee floor/bump logic.

To test this I've been performing an E2E test: Anvil with mining disabled, original poster tx, confirmation timeout, real same-nonce replacement, then resume mining and assert the receipt/nonce progression.

So, for this scenario, this PR fixes the fee-bump half of F3, but needs the gas-estimation adjustment for the poster to actually submit the replacement.

[stephenctw]

@jplgarcia Thanks for catching this — the fee bump was correct, but the real replacement path never reached eth_sendRawTransaction because Anvil rejected eth_estimateGas with the pending nonce (nonce too low).

Addressed by scoping the change to the same-nonce replacement path: estimate gas without that nonce, pin the result on the replacement tx, and keep the existing ≥10% fee floor/bump.

Also added an Anvil E2E along the lines of your repro: --no-mining → original poster tx → confirmation timeout → same-nonce replacement → mine → assert receipt/nonce progression (plus original eviction and on-wire fee floor).