[io] Avoid 32-bit overflow in ASCIIReader buffer sizing

[KoksalBerkay]

Fixes #6483.

ASCIIReader calculates buffer sizes using 32-bit multiplication before converting the result to std::size_t. On 64-bit systems, large inputs can therefore produce a truncated buffer size.

Promote one operand to std::size_t before multiplication in both the initial allocation and the final resize. This follows the approach approved in the issue and used by PLYReader.

Add a regression test that supplies synthetic large metadata through readHeader() and exercises the real read() method. The expected size exceeds vector::max_size(), allowing the test to check rejection without allocating a huge buffer.

Validation:

• The regression fails with the original implementation and passes with the fix.
• All 47 tests in the I/O executable pass.
• All 11 configured I/O CTest entries pass.
• git diff --check passes.

Tested on arm64 macOS with 64-bit std::size_t. This change does not add overflow checks for 32-bit systems.

[mvieth]

Thank you!