docs(adr): compress to 6 records in the domain-modeling format

[lesnik512]

Rewrites docs/adr/ from 13 records (~6,500 words) to 6 (~840 words), one paragraph each, in the shape org/docs/standard.md asks for. Survivors are renumbered 0001–0006 in their original order and every citation moves with them. Same treatment as modern-python/httpware#128.

Evaluation

Each record was tested against the admission test: hard to reverse, surprising without context, the result of a real trade-off. Records that fail it on their own but state a rule a survivor relies on were folded into that survivor as a sentence.

Old	Outcome	New
0001 healthcheck pass-through	kept	0001
0002 zero-dependency core	kept	0002
0003 network/namespace keys, 0004 stop_* inert, 0005 sysctls pod-level	merged: one rule, the shared namespace, decides all three classifications. 0005's "resolved" note and 0003's "dns deferred" were stale: both are pod-level and shipped.	0003
0006 no parse-don't-validate, 0007 graph queries recomputed	merged: one seam decision. 0006's long correction narrative collapses to the current truth, _plan calls validate itself.	0004
0008 no structural registry, 0010 no unified schema validator, 0013 volumes hand-rolled	merged: the same false-seam ruling at three sites. The 2026-07-15 blind reopen stays as one sentence.	0005
0009 docker-rejection parity, 0011 list-of-str refusals, 0012 negatives deferred	merged: 0011 and 0012 are worked examples of rule two and the host carve-out, now one clause each. The #87 residual stays.	0006

Citations

CONTEXT.md (structural key, rule one / rule two), compose2pod/graph.py, compose2pod/values.py, tests/test_cli.py, tests/test_values.py and tests/conformance/conftest.py all cited 0009 or 0008/0013 by full path and now cite 0006 or 0005. tests/test_adr_citations.py passes, and the repo-wide sweep finds no reference to an old or dropped number.

Checked: unit suite, ruff, eof-fixer, lychee --offline over the touched Markdown. Conformance and integration jobs run in CI.