docs(openkal): the design record in academic register, and current through 0.4

[Sunrisepeak]

The design document was written across four rounds and accumulated the register
of working notes: emphasis carried by symbols rather than by sentences, second
person, and colloquial phrasing. It is rewritten as declarative prose. No
technical claim is removed and no measurement is dropped; the changes are to how
the document reads, and to what it covers.

Three substantive additions accompany the rewrite.

A summary places the eight assessment dimensions at the head of the document,
each with the section that argues it: architecture, stability, simplicity,
consumer experience, compatibility, portability, consistency, and upgrade without
disruption. Previously the dimensions were argued in the text and nowhere
collected, so the document could not be read as an assessment.

The record now runs through 0.4 rather than stopping at 0.2. Three sections are
added: what the extension to eight interfaces derived and from what; the two
rules 0.4 states that 0.3 did not; and what a second implementation and a C
library above the stack establish — together with what they do not, since both
implementations were written by one author and their agreement on the argument
vector was both unanimous and wrong.

A section on consistency records that openkal was implemented twice without
modifying mcpp, and enumerates the mechanisms used: conditional dependencies,
platform-selected development dependencies, absence of a module as a compile-time
capability check, and a static comparison of the exported surface. None is new.
That every declaration a project makes resides in its mcpp.toml is the
constraint under which the capability configuration file was removed.

The completeness plan gains the dimension it was missing — consistency — and a
record of where the work stands against its own five criteria for completeness.
Two are met, one is partly met, one is not met and not attempted, and one is not
met for a reason worth stating: a second implementation by the same author is not
the second implementation that criterion asks for.