fix(ssestream): handle content-type parameters and case in decoder lookup

[MaxwellCalkin]

Summary

• NewDecoder performs an exact string match on the raw Content-Type header to find a registered decoder, but RegisterDecoder lowercases keys — creating a case-sensitivity mismatch
• Neither function strips media type parameters (e.g., ; charset=utf-8), so a Content-Type of application/vnd.amazon.eventstream; charset=utf-8 would fail the lookup
• When the Bedrock eventstream decoder is not matched, the binary eventstream response silently falls through to the text/event-stream SSE decoder, causing decode failures or data corruption
• The fix uses mime.ParseMediaType (Go stdlib) to properly extract and normalize the media type before decoder lookup, consistent with how requestconfig.go already handles Content-Type parsing elsewhere in this SDK

Before:

contentType := res.Header.Get("content-type")
if t, ok := decoderTypes[contentType]; ok {  // exact match, no normalization

After:

mediaType, _, _ := mime.ParseMediaType(res.Header.Get("content-type"))
if t, ok := decoderTypes[strings.ToLower(mediaType)]; ok {  // normalized + lowercased

Test plan

• Verify Bedrock streaming still works with standard content-type header
• Verify decoder is correctly matched when server returns content-type with parameters (e.g., application/vnd.amazon.eventstream; charset=utf-8)
• Verify decoder is correctly matched with mixed-case content-type headers
• Run existing test suite to confirm no regressions

🤖 Generated with Claude Code

AI Disclosure

This PR was authored by Claude Opus 4.6 (Anthropic), an AI agent operated by Maxwell Calkin (@MaxwellCalkin).