fix(sync): guarantee consecutive L1 gas price samples are fed into L1GasPriceProvider

[t00ts]

Relevant context for this PR can be found in #3197

---

This should now be way more reliable. Been running it for ~1d and it seems to work as expected.

Summary of key changes:

1. In our ethereum client:
   a) Made the get_gas_price_data_range strict. It used to just silently skip errors.
   b) Changed subscribe_block_headers from callback-based to channel-based. Otherwise I couldn't make the caller have full control over the processing loop.
   c) Also added tracking of parent block hash to L1GasPriceData to detect reorgs.

2. In the L1GasPriceProvider:
   a) Proper errors when adding a gas price sample. This allows the caller to take the right action accordingly.
   b) Detect reorgs by using those parent block hashes mentioned previously.

3. The new sync loop:
   There's an outer loop that runs forever (reconnects on failures) and an inner "cycle" which bootstraps the provider, subscribes to block header updates, and processes them. Here's where we pay attention to the two errors (gap or reorg detected) and we act accordingly. For gaps, we just inline fill them and keep going, for reorgs we restart the whole thing.

Worth noting that I changed the bootstrap initial block from finalized to latest. The latter is way closer to what we'll get on our first subscribed items, and thus reduces the initial gap fill. I think for gas prices it should be fine to rely on latest.

Closes #3196
Closes #3197

[vbar]

Worth noting that I changed the bootstrap initial block from finalized to latest. The latter is way closer to what we'll get on our first subscribed items, and thus reduces the initial gap fill. I think for gas prices it should be fine to rely on latest.

Note that this incompatible with the apollo implementation, which uses not just finalized blocks, but a configurable "finality gap" (but I agree the difference hopefully won't matter).