v0.19.0

Added

• A standard exports its rules, and a corpus adopting standards ships a skill that reads them. The type declares
  parts: over the H3 headings under ## Rules, so kac export writes one line per rule to
  standards/rules.jsonl, carrying the obligations in the markdown the standard wrote them in. Each record travels
  beside them with its Summary and its conformance checklist. kac bundle includes the new standards-lookup skill
  wherever the export carries standards, and trims it where it does not. kac new ships the skill in the plugin tree,
  and kac update sends it to a corpus already created.
• A field can be held to the citations its prose gathers. A type declares mirrors-citations: <Label> beside the
  field's ref:, and kac validate reports drift in both directions between the field and the labelled lines. A line
  is written in italic with the label bold, and closes the section whose citations it gathers, so a line standing in
  the middle of one is reported where it sits. A standard's implements: declares it: each rule closes on
  _**Covers:** …_ naming the clauses it discharges, so the frontmatter says which obligations the standard answers
  and each rule says which of them it answers. The obligations under a rule then carry no clause citation of their
  own, and kac export drops the footnote before it takes them, so a part carrying nothing else travels with no
  obligations rather than with a coverage line standing where its words belong. kac validate also reports the key
  declared with no ref: to resolve against. kac new ships the form in the standards template.
• A standard may carry a Sources and further reading section. It names the external documents the standard defers
  to, each marked normative or informative. A rule built on somebody else's conventions then says where the rest of it
  lives. kac new ships the section in the standards template, and a standard deferring to nothing deletes it.

Changed

• A standard's rules sit under ### headings, and kac validate reports a Rules section with none. The heading
  is what the rules beneath it hold a reader to, and it is the address a citation and an export both carry, so
  part-none now reaches standards as it already reached glossaries. A standard whose rules are a bare bullet list
  gains one heading. kac new ships the grouping in the standards template.
• A tool's category is the folder it sits in under tools/, as a policy's and a standard's already were. It was
  the one of the three still written by hand, and the only field in the schema carrying no description:. A record
  that writes the key now fails derived-key: delete the line and file the record under the folder you want it to
  name. It is no longer required, so a tool filed directly in tools/ simply has no category, which is what the other
  two do. kac new ships the template without the key.

Fixed

• kac validate reports a schema declaring of: object with no entry: block, rather than ending in a stack
  trace. The schema pass already names that fault as schema-shape. A record filling such a field reached the
  entry check first and took the run down, so the message never printed.
• Declining kac update --drop-type says update, where it used to say new. The message is the tail of the
  command that printed it, and the one it named was a command the reader had not run.
• A rule whose expr: names a number too large for a whole number is reported, rather than ending in a stack
  trace. words() < 99999999999 reached the parser's integer literal and overflowed past the exception the schema
  load catches. kac validate now names the number, its position and the rule.
• kac restore refuses a package that unpacks to more than 256MB, or that holds a single entry over 16MB. The
  path each entry names was already held inside the import folder, and what it unpacks to was not, so a malformed
  package was read whole into memory instead. Both caps count the bytes actually read, because a zip entry's
  declared size is the package's own claim about itself.
• A part id written against a link takes no delimiter that closes an emphasis. _[pol-SCRT].EMBED_ read the
  citation as pol-SCRT.EMBED_, because the id was measured off the source rather than off the text markdown makes
  of it. part-ref reported a clause nobody could write.