Reject oversized --parameter-scan ranges without overflowing i32

[00200200]

--parameter-scan panics in debug builds (and wraps in release) when the range is so wide that (end - start + 1) / step overflows i32. That happens while computing the size hint, so the existing 100,000-parameter cap never gets a chance to reject the range.

$ hyperfine --parameter-scan x -2147483648 2147483647 true
thread 'main' panicked at src/parameter/range_step.rs:85:18:
attempt to subtract with overflow

A narrower but still huge span overflows on the + 1 instead:

$ hyperfine --parameter-scan x 0 2147483647 true
thread 'main' panicked at src/parameter/range_step.rs:85:17:
attempt to add with overflow

Integer scans now compute that count in i64 and return the usual "Parameter range is too large" error.

Related to #920. The next() overflow near i32::MAX is covered by #921; this is the remaining construction-time overflow called out in that review.

On master the two new unit tests panic with attempt to add/subtract with overflow. With this change they pass, along with the CLI test parameter_scan_full_i32_range_is_rejected. cargo test: 46 + 19 + 40.