fix: reject out-of-range integers on write and search instead of silently truncating

[JeanExtreme002]

Summary

A fixed-width ctypes integer setter silently wraps an out-of-range value
(e.g. 2**40 into a 4-byte slot stores 0). As a result:

• write_int(addr, 2**40) reported success while actually writing 0 —
  silent corruption of the target's memory.
• search_by_value(int, value=2**40) encoded the target as 0 and quietly
  matched every zeroed slot in memory instead of erroring.
• The unsigned write helpers (write_uint, …) raised, but with int.to_bytes'
  cryptic OverflowError ("int too big to convert") instead of a clear error.

This adds a single range check (_check_int_fits) applied at every numeric
coercion point, so an out-of-range integer fails fast with an actionable
ValueError everywhere:

Coercion point	Path
prepare_write	signed / generic write (write_int, write_short, …)
AbstractProcess._write_unsigned	unsigned helpers (write_uint, …)
value_to_bytes	search-target encoder (search_by_value / _between)
RemotePointer.value setter	inherited via write_process_memory

The accepted window is the union of the signed and unsigned ranges for the
requested width, so a value in either representation stays valid (0xFFFFFFFF
in a 4-byte field is the bit pattern of -1); only values that genuinely
cannot be encoded in that width are rejected.

Also included

• Regression tests for every path above, plus a test pinning that a write
  to an unmapped address still raises OSError.
• NOT_VALUE_BETWEEN scan coverage (typed-int fast path + bytewise
  fallback), which previously had implementations but no test.
• Functional tests for the Qt app's value_types parse/format layer and
  the CheatEntry JSON round-trip (previously untested pure logic).
• CONTRIBUTING: route the dev commands through the existing Makefile targets
  and fix the stale note claiming macOS is excluded from CI.

Testing

Full suite green, including the slow self-process integration tests:
428 passed, flake8 clean, mypy PyMemoryEditor clean.