Add input-reader module

[stephenctw]

Add safe-input-reader and related storage logic

[GCdePaula]

Nice work!

I’ve left two inline comments (range split + atomic cursor/input persistence).

There are two follow-ups I suggest we handle on top of my open branch:

1. Input reader lifecycle
   Right now the reader thread is detached, so we don’t have proper shutdown/join/error propagation to main.
   In my open PR we already have a cleaner lifecycle pattern for background workers; I suggest we wire the reader into that same pattern.

2. Config surface
   This PR introduces new env knobs.
   My branch has moved config parsing to clap, so we should add the new reader options there to keep configuration consistent.

Given that, my recommendation is:

• merge my branch first,
• then rebase this input-reader work on top,
• and apply the two inline fixes plus the two follow-ups above.

[GCdePaula]

We're not handling the base case. Moreover, the middle calculation can underflow.

I propose these changes:

if start_block >= end_block {
    // Cannot partition further; bubble original error.
    return Err(vec![e]);
}

let middle = start_block + (end_block - start_block) / 2;

[GCdePaula]

These two operations have to be atomic; we need to perform them inside a single db transaction.

I'd create a single function in storage called something like append_safe_inputs_and_advance_cursor(inputs, new_last_processed_block) that performs both operations atomically.

[stephenctw]

Given that, my recommendation is:

• merge my branch first,
• then rebase this input-reader work on top,
• and apply the two inline fixes plus the two follow-ups above.

Sounds good, let's merge your PR first, then adjust mine afterwards.