docs: preset facts outside docdrift's scope are stale — wrong default preset in 5 places, stale README pipeline lists

[OsherElhadad]

The docdrift guard added in #142 covers two files (docs/how-to/choose-a-preset.md,
docs/reference/presets.md). Preset facts stated outside those two files are stale, and
the guard's shape cannot reach them.

1. The default preset is house; five sites say codesmart

cmd/context-guru-proxy/main.go:46:

preset = flag.String("preset", envOr("PRESET", "house"), "... house = the service default ...")

Sites that say otherwise:

Location	Says
README.md:111	# --preset codesmart; listens on :4000
README.md:132	Presets: **`codesmart`** (the default — the SWE-bench-winning cache-aware config
README.md:143	flag table: ``
docs/reference/config.md:79	``
docs/get-started/quickstart-proxy.md:20	# default preset: codesmart

house is [format, dedup, toon, cmdfilter, searchfold, textclean, extract, cachesplit, toolfilter]
— different economics from codesmart, and it still runs toon, which is retired from every
preset. Anyone who runs the binary bare while reading any of these five is measuring a
different config than the one the published SWE-bench numbers describe.

quickstart-proxy.md:20 is the worst of the five: it is step 2 of the first page a new user
runs.

2. README.md:132-136's pipeline lists are both stale

codesmart  doc  [format, toon, dedup, failed_run, cmdfilter, extract_llm, extract, cachesplit]
           code [format, textclean, searchfold, dedup, failed_run, cmdfilter, extract_llm, extract, linecap, cachesplit]
codesafe   doc  [format, dedup, failed_run, cmdfilter, extract, collapse, cachesplit]
           code [format, textclean, searchfold, dedup, failed_run, cmdfilter, extract, collapse, linecap, cachesplit]

The trailing name list also omits agentdiet, house, and housellm.

3. docs/components.md:34-56 claims "verbatim" and is not

12 rows, all currently correct, but it omits house/housellm while claiming to reproduce
Presets (`config/config.go`), **verbatim**.

Why the guard cannot cover these

docRow is anchored ^\|\s* — it matches markdown table rows only. README's and
components.md's lists are prose, and config.md:79's first cell is
`--preset` / `PRESET`, which the pattern does not match either. Widening the guard to
these files matches zero rows.

Suggested fix

Delete the duplicates rather than guard them. config/form.go and dash/ui/index.html:67
already do this correctly — they derive from config.PresetNames() and from runtime data, so
they cannot rot. Replace README's and components.md's bracket lists with a pointer to
docs/reference/presets.md, and have the default-preset claims say "see --help" or be
generated from the flag.

Found during a review sweep of #129/#130/#136/#137/#138/#141/#142 against main e88f5d8.

[amiddavid]

Fixed by #161 (merged).

All five wrong-default sites now say house: README's run line, its preset prose and its flag table, docs/reference/config.md:79, and docs/get-started/quickstart-proxy.md. The two config/config.go comments that asserted codesmart is the proxy default are corrected too — those were the source, sitting three lines from the flag that disproves them, which is how the claim reached five documents.

README's codesmart/codesafe pipeline lists are regenerated from the presets map, so toon is gone from both.

Verified on main after the merge:

• README.md:134 — "house is the binary's default. codesmart is the SWE-bench-winning…"
• docs/reference/config.md:79 — | --preset / PRESET | house | … codesmart is the SWE-bench arm and must be asked for by name. |

Not covered by #142's drift guard, which reads only the two preset tables — that gap is what this issue reported.