perf(puffin): reuse prefetched footer payload

[fallintoplace]

What

• Reuse the footer payload from the initial 8 KiB tail read when the whole footer fits.
• Keep the existing io.SectionReader fallback for larger footers.
• Add read-count coverage for both paths.

Why

readFooter already reads the tail to avoid extra object-store round trips. Small footers were still decoded from a section reader backed by the underlying ReaderAt, so the payload was read again.

Performance

With a small footer and a simulated 5 ms delay for every range read:

• Before: 17.8 ms/op
• After: 12.2 ms/op
• About 1.5x faster

This removes one redundant range read from the small-footer path.

The benchmark used 5 runs of 20 iterations on an Apple M1 Pro. The exact numbers will vary with object-store latency.

Tests

• go test ./puffin
• go test -race ./puffin
• go test ./...