fix(slice): explicit zero end no longer returns the whole array

[spokodev]

Problem

A slice with an explicit 0 end returns the entire array instead of an empty result:

JSONPath({ path: '$[:0]',  json: [0,1,2,3] }) // → [0,1,2,3]   (expected [])
JSONPath({ path: '$[2:0]', json: [0,1,2,3] }) // → [2,3]       (expected [])
JSONPath({ path: '$[0:0]', json: [0,1,2,3] }) // → [0,1,2,3]   (expected [])

For comparison, $[:1] → [0] and $[2:1] → [] are already correct — only an explicit 0 end is mishandled. Per the Python slice semantics this library documents, arr[:0] is [].

Cause

In _slice the end index is computed as:

end = (parts[1] && Number.parseInt(parts[1])) || len;

When parts[1] is '0', Number.parseInt('0') is 0, which is falsy, so || len replaces the legitimate index 0 with the array length.

The slice loc is gated upstream by /^(-?\d*):(-?\d*):?(\d*)$/, so parts[1] is always either an empty string (end omitted) or a valid integer string. The fix defaults an omitted (empty) end to len and parses any explicit value, including 0:

end = parts[1] ? Number.parseInt(parts[1]) : len;

start and step are intentionally left unchanged: start's falsy default is already 0, and step's || 1 is a deliberate guard against an infinite i += 0 loop.

This is the same 0-is-falsy class as the @property fix in #245.

Verification

• New regression test in test/test.slice.js ($[:0], $[2:0], $[0:0] → []); fails before, passes after.
• Full suite: 278 passing.
• Differential vs Python slice semantics over every valid start:end:step shape on arrays of length 0–7 (4,608 cases): 264 divergences before → 0 after. All 264 were end == 0 cases; the fix introduces no new divergence.