REQ-DEM-062: which figure should the EVM row hold for an offset format?

[TGoodhew]

The question

REQ-DEM-062 asks for an Offset EVM variant and states its acceptance against a contrast:

on a clean OQPSK signal returns near-zero where conventional EVM computed at a common instant
does not — the difference between the two is what shows the half-symbol stagger is honoured.

That requires both numbers to exist. It does not say which of them the plain EVM row should
hold, and the two readings differ by a factor of nearly two thousand:

OQPSK, clean, 512 symbols:
  Offset EVM   0.028734 %rms   (I at the symbol instant, Q half a symbol later)
  EVM         55.2910   %rms   (both parts at the one instant)

What was implemented

• Offset EVM = the staggered reading, present only for offset formats (absent, not NAN, for
  the rest — MetricApplicability already draws that distinction).
• EVM = the common-instant reading, which is REQ-DEM-060's formula applied literally at one
  decision instant and is what an analyser that did not know about the stagger would report.
• DemodResult.EvmPercent — the headline, and what the lock diagnosis, the shell and every
  existing test read — = the Offset EVM for offset formats. Unchanged behaviour.

The reasoning for the split: showing only the staggered figure would be showing a good number with
no way to tell whether it is good because the signal is or because the measurement was told what it
wanted to hear. Showing only the common-instant figure would call a perfectly good OQPSK
measurement a failure. Showing both, side by side, is self-explanatory — and the ratio is itself a
useful diagnostic.

The cost is that an offset format's summary has an EVM row reading tens of per cent while the
headline reads hundredths of one, and a user who reads the row without the row beneath it will be
misled.

What might be decided instead

1. As implemented. Two rows, two meanings, the ratio visible.
2. EVM mirrors Offset EVM for offset formats — the two rows read the same, which is what
   several instruments of this family do, and the contrast lives only in the test. Simpler to read,
   but the EVM row then carries no information the row below does not.
3. Rename the rows so the common-instant figure is labelled as what it is — but REQ-UI-053
   fixes the label set and EVM is in it, so this needs a specification change.

My recommendation is (1) with the shell stating the distinction next to the table, which is what
the normalisation line added for REQ-DEM-061 already does for its own ambiguity. If the row is
ever seen without that line, (2) is the safer default.

To unroll to (2): ErrorMetricStep writes both rows in one place; delete the second Replace call
and the CommonInstantSymbols projection in JointRefinementStep, and adjust
OffsetEvmTests.TheHeadlineEvmIsTheOffsetOneAndNotTheCommonInstantOne.

Refs #170.