feat: frame content size in encoder output

[polaz]

Summary

The encoder currently sets frame_content_size: None in the frame header (frame_compressor.rs:145). The C reference defaults to writing content size when known — this enables decoders to pre-allocate output buffers, significantly improving decompression speed.

C reference behavior

• ZSTD_c_contentSizeFlag defaults to 1 (enabled)
• When pledged source size is known, it's written in 1/2/4/8 bytes depending on value
• Single-segment frames (content fits in window) use content size instead of window descriptor — saves 1 byte per frame

Current Rust state

• frame_header.rs:74 — FCS field always None
• Header always writes window descriptor even when content size would suffice
• frame_compressor.rs:144-148 — single_segment: false always

What needs to be implemented

1. Content size tracking — count total bytes read during compression
2. Two-pass or buffered approach — either buffer all input to know size upfront, or write placeholder and seek back
3. Single-segment optimization — when content ≤ window, set single_segment: true and omit window descriptor
4. FCS encoding — 1/2/4/8 byte encoding based on value

Acceptance criteria

• Frame content size written when source size is known
• Single-segment flag set when applicable
• C zstd decompresses correctly (validates FCS)
• Decompressor can pre-allocate buffer from FCS

Time estimate

1d