fix: update recommended memory for ops setup

fix: edge case in go/txmgr not waiting for numConfs

feat: add teleportr Postgres backend

feat: update message-relayer to use the SDK

feat(core-utils): remove Watcher and injectL2Context

An issue was recently reported (#2191) that showed that 4gb was not
sufficient to run the docker setup. This commit updates the recommended
amount of memory to 8gb.

Replaces usage of injectL2Context with asL2Provider within the batch
submitter and the replica healthcheck service.

Removed duplicated utilities in data-transport-layer

maintenance: minor README updates

Removes all message relaying utilities found inside the message-relayer
package. All of these utilities are now present in the SDK. Also
re-implements the message-relayer service with the utilities found in
the SDK.

This change adds an additional IsWaitingForConfirmation method to the
txmgr.SendState, which returns true if it is aware of at least one mined
tx. This allows the txmgr to skip publishing new txs while we are
waiting for additional confirmation on a mined tx, which in turn reduces
the number of useless API calls we make to the backend provider. As an
additional benefit, this also reduces the likelihood of a false positive
sample from ShouldAbortImmediately since it will stop processing errors
while we have a mined tx.

This commit adds a new SendState struct, which is used internally by the
txmgr to determine if we should give up on a particular tx at the given
nonce due to ErrNonceTooLow failures. To do so, we track the set of
mined txns at any particular time, which are updated via asynchronous
calls to TxMined and TxNotMined by each of the spawned goroutines. The
presence of a mined txn supercedes any of the errors returned, and will
cause the txmgr to wait for full confirmation. However, if we observe
multiple ErrNonceTooLows, the SendState will wait for a configurable
safe abort count to ensure that the reading wasn't due to being in a
transient state, as we don't have exact guarantees on the execution
ordering of various goroutines.

With this change, the test added in the prior commit now succeeds.

This commit adds a test case for the following scenario:
 1. The first few publications succeed with err == nil.
 2. One of the txs is mined.
 3. Subsequent publications start failing because they are using the
    same nonce as the tx mined in (2).

This edge case is currently overlooked in the txmgr, and will cause the
shouldAbortImmediately method to return true and not wait for the
appropriate confirmations.

Rather than having a combined update the gas price and publish the tx,
we split this this into two methods: UpdateGasPrice and SendTransaction.
This will allow the txmgr to obtain the txHash before publishing the
tx.

Version Packages

Develop -> Master

fix: skip an unreliable integration test

Skips the unreliable timestamp/blocknumber test which would often be off
by one or two. Creates an annoying habit of having to rerun the
integration tests until the test passes, which isn't how integration
tests should work.

go/proxyd: Handle nil responses better

fix: use correct token for label-requires-review

This check is the only one we have that uses PERSONAL_ACCESS_TOKEN,
perhaps it's expired or not present.

fix: make relayer work again

fix: remove extra space in BSS metric names

Fixes a bug where the relayer package was not correctly running the
start script since the start script updated recently.

feat: unify BSS metric name format, switch num_elements_per_batch to Summary

bss: properly set gas limit on clearing txs

feat: add LOG_TERMINAL flag, default to JSON

test(integration): replace watcher with sdk CrossChainMessenger

The histogram doesn't seem to render too well on Grafana, but the
batch_size_bytes one does. This commit modifies the
num_elements_per_batch to also use a Summary with the same buckets.