Expose CheckAttempt to TxManager, saves some unnecessary calls

[j16r]

The initial phase of the EthTx adapter creates a transaction, then attempts to do gas bumping on it.

Gas Bumping is never needed on the initial transaction so half of this code is redundant, and because the existing records/information is not pushed into the BumpGasUntilSafe method, which is meant to be used with a freshly retrieved Tx, it creates a lot of redundant work. BumpGasUntilSafe actually needing to do any work in real life is highly unlikely, so I have optimised this path: instead just check for a receipt on the odd chance that the chain was quick to accept.

Full list of changes:

• CheckAttempt has no side effects, and clearly indicates the states of a TxAttempt, this can now more easily be unit tested.
• processAttempt (previously checkAttempt) now does all the work of gas bumping retrying etc. The logic of this code is far easier to understand now because the logic for determining state is no peppered in together.
• BumpGasUntilSafe/CheckAttempt always return a receipt if one was retrieved, this makes them safer to use - the logic for using them can be simplified
• BumpGasUntilSafe also now returns a state, and always returns a receipt (because it uses CheckAttempt which does the same) which means we can distinguish between fatal errors and just receipt retrieval errors, which we choose to ignore so that attempts can be checked again for the next head

[dimroc]

Optional

This line originally threw me off, because conceptually it's impossible for something to be safe before something that's sent. I later discovered you're using the receipt's BlockNumber to gauge whether something is safe so this works out.

If possible, I would recommend renaming sentAt to something else, perhaps currentHeight?

I would also rename the test TestEthTxAdapter_Perform_ConfirmedWithBytesAndNoDataPrefix from using the language Confirmed to using the more appropriate language Safe: TestEthTxAdapter_Perform_SafeWithBytesAndNoDataPrefix