Document native batch and parallel decoding in the Guppy DEM workflow

[ciaranra]

Summary

The Guppy DEM decoding workflow doc previously showed per-shot Python decode loops and mentioned decode_count in one sentence. This adds a section on the native SampleBatch decode paths:

• decode_count, decode_count_parallel, and decode_count_batch, with guidance on when each pays off.
• The configuration caveat that type strings construct decoders with default settings, including the "pymatching"-is-correlated vs PyMatchingDecoder.from_dem-is-uncorrelated distinction (the executable blocks use "pymatching_uncorrelated" so their equality assertions against the hand-built loop are exact).
• Wrapping simulated shots from evaluate_result_columns() in a SampleBatch so stage 4b shots ride the same native parallel path.
• A pointer to decode_stats_parallel for latency quantiles.

All new code blocks are tested: the doc-test suite (just docs-test) passes with the generated workflow test running all 11 blocks, and the equality assertions between the native paths and the hand-written loops execute as part of it.

Verification

• just docs-test: 288 passed, 12 skipped (includes the 11 generated blocks for this doc).
• just lint: clean, no formatter rewrites.