fix(fetchers): harden youtube transcript handling

[chaliy]

Motivation

• The YouTube timedtext transcript path previously fetched and built unbounded transcript text and then truncated with a byte-indexed slice, allowing excessive memory use and a panic on UTF-8 multibyte boundaries which can cause a denial-of-service.

Description

• Pass options.max_body_size into fetch_transcript(...) and bail out when the timedtext XML exceeds the configured cap to avoid unbounded memory use.
• Introduce MAX_TRANSCRIPT_CHARS and replace unsafe byte-slicing truncation with a safe_truncate_utf8 helper to guarantee UTF-8-safe truncation in format_youtube_response.
• Keep transcript parsing logic but enforce the size check before joining segments and return None on oversized payloads.
• Add a regression unit test test_safe_truncate_utf8_multibyte_boundary that validates truncation does not cut inside a multibyte code point.

Testing

• Ran cargo fmt --all to format changes successfully.
• Ran cargo test -p fetchkit test_format_youtube_response_truncates_long_transcript and it passed.
• Ran cargo test -p fetchkit test_safe_truncate_utf8_multibyte_boundary and it passed.

---

Codex Task