Change BlockLocator index selection logic

[greymistcube]

Given hashes [0x00, 0x01, ..., 0x0e, 0x0f], we should expect the following results:

• BlockLocator created with sampleAfter value 0: [0x0f, 0x0e, 0x0c, 0x08, 0x00]
• BlockLocator created with sampleAfter value 1: [0x0f, 0x0e, 0x0d, 0x0b, 0x07, 0x00]
  • Current actual value: [0x0f, 0x0e, 0x0c, 0x08, 0x00]
• BlockLocator created with sampleAfter value 2: [0x0f, 0x0e, 0x0d, 0x0c, 0x0a, 0x06, 0x00]
  • Current actual value: [0x0f, 0x0e, 0x0d, 0x0b, 0x07, 0x00]

In the last example above, [0x0f, 0x0e] would correspond to non-sampled indices and and [0x0d, 0x0c, 0x0a, 0x06, 0x00] would correspond to sampled indices. From my perspective, there are two problems with current implementation:

• There is no difference between BlockLocator created with sampleAfter value of 0 and 1.
• The initial delta of indices for "sampling" starts with 2 not 1.

Given starting index $n$, I think the cleanest way to define the act of "sampling indices" would be the following:

• $i_{0} = n$
• $i_{1} = i_{0} - 1$
• $i_{k} = i_{k-1} - 2(i_{k - 2} - i_{k - 1})$ for $k \geq 2$

Or in a more compact form:

• $i_{0} = n$
• $i_{k} = i_{k - 1} - d_{k - 1}$ where $d_{t} = 2^{t}$ for $k \geq 1$

Or in a closed form:

• $i_{k} = n - (2^{k} - 1)$