New scanner decoder with much higher performance

[CarlLindqvist]

Scan decoder vs original tokenizer decoder:

• 2.4x faster (170 us → 70 us)
• 53% less memory (58 KB → 27 KB)
• 53% fewer allocations (776 → 368)

Scan decoder vs encoding/xml:

• 10.8x faster
• 10.3x less memory
• 18.2x fewer allocations

[Lunkers]

LGTM overall, but I'd like some clarification on one thing. If we're doing unsafe pointer modifications for all the vmap strings we might run into issues here; no guarantees that the manifest URLs will fit within the original creative URLs