bss: split up SubmitBatchTx method in Driver iface

Previously it was being logged every time we modified the gas price.
Additionally, we log the raw byte length rather than tx.Size(). The
latter returns a float value of type StorageSize, which explains why the
metrics haven't been indicating actual byte values.

The current interface supports only one method for both crafting and
publishing a batch transaction, SubmitBatchTx. Currently this is being
executed on each new gas price that the txmgr commands, implying we are
doing a lot of extra work to rederive batches. In addition, much our
instrumentation lives inside this method, meaning that the they are also
being recorded multiple times per transaction. When we get to
processing larger batches on Kovan and Mainnet, this could also become a
resource bottleneck.

This commit remedies all of the above issues by splitting out the
transaction crafting process from the publication. A new method,
CraftBatchTx, is added to the Driver interface. This method is
responsible for creating a fully formed transaction, but does not
publish it. The responsibility of SubmitBatchTx is now modified to take
an predefined transaction, i.e. created by CraftBatchTx, overwrite the
supplied gas price, and publish. In this way, this expensive call to
build batches can be done once before handing the transaction of the
txmgr.

fix(batch-submitter): clear state root batches

Fixes a bug where both the sequencer and proposer main loops were
attempting to clear the pending transactions from the batch-tx
submitter's wallet. The impact is that we may have been overspending on
fees due to conflicting/reverting state batches. However the impact
overall should be minor given the relative size of state batches in
comparison to tx-batches.

feat: implement clear pending txs for go batch submitter

feat: L2Provider test scaffold

feat(proxyd): Support immutable RPC caching 

Makes sure that the Default sender is address(0)

feat: add BatchPruneCount metric

This metric gives insight into how many times a batch is pruned to meet
the required MaxL1TxSize.

go/proxyd: Add request/response payload size metrics

l2geth: update timestamp logic

Update the assertion to not expect that the timestamps are
exactly equal but instead assert that the values are within
5%. This is because now the timestamps in L2 no longer
correspond to L1 timestamps and may be updated arbitrarily
by the sequencer.

This commit updates the timestamp updating logic
such that `time.Now` is used instead of relying on
L1 timestamps. This gives a higher fidelity for the
`TIMESTAMP` opcode as well as makes the time on L2
be closer to the time on L1.

L1 to L2 transactions no longer have the property of
having the same timestamp on L2 as the timestamp
of the L1 block they were included in.

This should be able to be turned on without needing
hardfork logic as replicas should always accept the
timestamp that the sequencer sets. The sequencer is
a trusted entity in the existing implementation and
it is expected that the sequencer will become more
trustless in future iterations of the protocol.

This change is added to improve both UX and devex.
Users are confused by the timestamps on Etherscan
being ~15 minutes behind. This is due to the timestamps
being set based on L1 block numbers, and the system
only pulls L1 data once a secure amount of PoW
is placed on top. Developers would like the timestamps
to have a higher fidelity and be closer to the timestamps
on L1.

go/batch-submitter: Metrics updates

- Renames a couple of metrics to include their units.
- Adds histogram buckets to `num_elements_per_batch`.
- Changes `batch_size_bytes` to a summary.
- Changes a couple of metrics to be gauges rather than counters.