docs(v2): record the standard-coverage debt as Tier 5

[s2x]

Seven pieces of standard that a generator in the Shipped table does not
implement were tracked only in conversation. This makes them a tier.

They are collected rather than filed against their own entries because they
share a shape: each encoder was finished against what the round trip could
prove, and these are what the round trip never asked about. None of them is a
wrong symbol — each either costs modules a standard-complete encoder would not
spend, or refuses a payload the symbology could have carried.

Ordered by what it costs to leave alone:

1. FfiEncoder intermittently reads a garbage size and the suite segfaults (present on main) #201 — the extension pass segfaults on roughly half of all CI runs and
   is continue-on-error: true, so a real extension regression would look
   exactly like the crash that is already there. The only defect on the list
   and the only one that hides other failures. Undiagnosed.
2. QR chooses no mode — Encoder.php passes Mode::Byte twice and nothing
   else, while DataEncoder implements numeric, alphanumeric and Kanji
   compaction that nothing reaches. Widest gap in ordinary use: eight bits a
   digit where the standard spends three and a third. The work is
   segmentation, not compaction.
3. Data Matrix is ASCII only — no C40, Text, X12, EDIFACT or Base256.
   Digits are already optimal; text costs a third more than it should and
   binary goes through ASCII upshift.
4. PDF417 has no ECI header and no macro block — the macro block is the
   one with a capability behind it: without it a payload over one symbol does
   not fit rather than taking two.
5. Aztec has no FLG(n) — the one entry whose fix adds a symbology rather
   than shrinking a symbol. FLG(0) is FNC1, so gs1-aztec is that plus the
   FNC1 placement rule the GS1 layer already has.
6. MaxiCode mode 5 — kept open for the reason Tier 3 already gave: its
   interleaving is not the plain mode's, nothing available writes one, and
   thirty candidate splits produced nothing the reader accepts.
7. DataBar Expanded's fourteen variable-measure compaction methods — last
   on purpose. Narrowest set of inputs, fourteen field layouts to get wrong,
   one or two symbol characters saved.

Tier 6's opening paragraph claimed 5 was being left free for a symbology tier.
It now says what 5 holds.

Every claim in the new section was checked against the source rather than
recalled: Encoder.php:145/:222 for the QR modes, the absence of
Mode::Numeric/Alphanumeric/Kanji anywhere outside DataEncoder, the
AsciiEncodation docblock, the "not implemented" notes already in
Encoding\Aztec\HighLevelEncoder and Encoding\Pdf417\HighLevelEncoder, the
missing case 5 in Encoding\MaxiCode\Mode, and ci.yml:194-200.

Docs only — no source, tests or fixtures touched. composer lint:kb ok.