New Marshal functions for VAST and VMAP

[CarlLindqvist]

VMAP marshaling (testVmap.xml — 3 ad breaks with full VAST):

	ns/op	B/op	allocs/op
xml.Marshal	~218,000	205,002	994
MarshalVmap	~57,500	150,144	6
Speedup	~3.8x faster	27% less memory	99.4% fewer allocs

VAST marshaling (testVast.xml — 2 ads with full creatives):

	ns/op	B/op	allocs/op
xml.Marshal	~22,500	24,936	115
MarshalVast	~3,930	5,376	1
Speedup	~5.7x faster	78% less memory	99.1% fewer allocs

Approach

The encoder in encoder_fast.go avoids all reflection by writing directly to a []byte buffer with append. Key optimizations:

• No reflection — hardcoded struct field traversal matching xml.Marshal field order
• Single allocation — one pre-allocated buffer, grown as needed (1 alloc for VAST, 6 for VMAP due to buffer regrowth)
• Inline Duration/TimeOffset formatting — avoids fmt.Sprintf and strings.Builder allocations
• Chunk-based entity encoding — copies runs of safe bytes at once rather than byte-by-byte

[Lunkers]

LGTM