fix: size the argv buffer in UTF-8 bytes, not UTF-16 code units

[madeiragab]

args_get writes each argument with TextEncoder, so the bytes that reach guest memory are UTF-8. args_sizes_get computes the buffer size with arg.length, which counts UTF-16 code units.

For ASCII the two agree and nothing happens. A single non-ASCII character makes the reported size smaller than what args_get actually writes, so the write runs past the end of the guest's argv buffer and corrupts whatever follows it.

environ_sizes_get, a few lines below, already measures with TextEncoder. This makes the argv path match it.

How I ran into it: a Rust database compiled to wasm32-wasip1, driven from the page, taking SQL as a command-line argument. One accented character inside a SQL comment corrupted the next argument in guest memory - it surfaced as the engine complaining about a column named C that nobody had written. Passing the script through stdin instead of argv worked around it on my side, which is what pointed at the argv path.