Decompression bomb in `ruzstd`: a tiny input can exhaust memory (OOM)

Summary

A crafted `.zst` input under 128 KB can force `ruzstd` to allocate several gigabytes at once, crashing the process with an out-of-memory condition. Since `ruzstd` is a decoder for untrusted input, a single small file is enough to take down anything decoding it.

The compressed size of a block was capped at 128 KB, but a compressed block's **decompressed** size was never checked against `Block_Maximum_Size = min(Window_Size, 128 KiB)` (RFC 8878). In `execute_sequences` the output was built with no running bound and no flush between blocks, so one block could expand to gigabytes — bypassing both the configured window-size limit and the block-by-block streaming that would otherwise release memory. (A related `u32` counter overflow on such a block caused an additional panic.)

This is a denial of service only; memory access stays in bounds (no corruption, no RCE).

`ruzstd` is also built as a dependency of the Rust standard library via `rustc-dep-of-std`; in the toolchain it is used only to load zstd-compressed debug symbols, so the toolchain-side impact is limited.

**Fixed in 0.9.1** (commit `d3907c9`), which enforces the per-block output bound. Upgrade to `ruzstd >= 0.9.1`.