Handle missing datum strips and correct RC6 slow tests

[pzfreo]

Fixes #2182.

Change

Only rank coincident datum stems when their requested view side has a strip. A missing strip now reaches the existing corridor no-strip path, which runs the normal legal fallback and records pmi_dropped if the datum cannot be placed. Imported surface-normal datums remain restricted to their proven side.

Reproduction and checks

• A public build_drawing(..., pmi="annotate") regression declares an imported side-view datum on the absent left strip and a separate PMI control frame. On unchanged main it raises the reported AttributeError at _gdt.py:792.
• With this patch the drawing completes, the other PMI frame remains, the missing datum has a pmi_dropped issue, and the solve trace proves the left strip was None and the datum's outcome was no_strip.
• All 47 tests in test_gdt_placement.py and test_strip_zones.py passed. scripts/pr-check --static and git diff --check passed.
• The reporter's AP242 file is unavailable locally, so this verifies the exact failing renderer state through the public build path rather than claiming a replay of that file.

RC6 slow-gate correction

The post-merge main slow run for #2181 had two stale expectations:

• TestTheCaseStudy still expected the automatic plate drawing to use A1. On current main, measured through build_drawing, both the automatic and explicitly two-principal-view builds use A2 at 1:1, have the same three views and 34 annotation names, and clean lint. The test now checks automatic/manual parity instead of the obsolete A1/A2 contrast.
• The source-owned GRM-03 manufacturing test still expected automatic A4 at 2:1 with a detail. Current main selects A4 at 5:1 with no detail, keeps the manufacturing row and its three source bindings, and has no error or dropped-requirement lint. Its declared Sheet replay still chooses A4 at 2:1 with a placed detail and retains the manufacturing meaning; the test checks those two observed arrangements separately.

Both isolated slow nodes and static checks pass locally. These are test corrections for #2181's already merged behavior; they change no production layout code.

Architecture

ADR 1 keeps the same compiler and annotation pass. ADR 2 retains the shared corridor solver and its existing no-strip fallback. ADR 3 recognition and ADR 4 authored meaning are unchanged. ADR 5 reports an unplaceable sourced datum while keeping other PMI instead of aborting the build. No ADR boundary or invariant changes.

ADR 2 still contains the previously logged contradiction between its no-automatic-principal-removal sentence and the current thin-plate behavior. The record itself is unchanged pending maintainer sign-off; #1130 tracks it.