LEAPR discre: idone flag reuse drops all but the first in-range discrete-oscillator delta line (twt = 0) #402

In src/leapr.f90, the delta-function block of discre (the twt <= 0
branch, lines ~1549–1585 in NJOY2016.78) reuses the idone flag for two
different loops — the outer loop over discrete lines and the inner
nearest-beta grid search:

      !--add the delta functions to the scattering law
      !--delta(0.) is saved for the incoherent elastic
      if (twt.le.zero) then
         m=0
         idone=0
         do while (m.lt.n.and.idone.eq.0)      ! outer: lines
            m=m+1
            ...
               if (bes(m).lt.zero) then
                  be=-bes(m)
                  if (be.le.betan(nbeta-1)) then
                     db=1000
                     idone=0
                     j=0
                     do while (j.lt.nbeta.and.idone.eq.0)   ! inner: grid search
                        j=j+1
                        jj=j
                        if (abs(be-betan(j)).gt.db) then
                           idone=1                          ! <-- also exits the OUTER loop
                        else
                           db=abs(be-betan(j))
                        endif
                     enddo
                     ...
                     if (add.ge.tiny) sexpb(jj-1)=sexpb(jj-1)+add
                  endif
               endif
         enddo
      endif

When the inner search finds the nearest grid point it sets idone = 1,
which also terminates the outer do while (m.lt.n.and.idone.eq.0)
loop. The net behavior: only the first in-range negative delta line is
added to the scattering law, and every subsequent line is silently skipped.
Because the lines were weight-sorted just above (the "sort the discrete
lines" block), "first" means the largest-weight in-range line — the rest of
the oscillator delta weight never reaches S(α,β).

Trigger: any deck with twt = 0 and discrete oscillators whose line
spectrum has two or more in-range downscatter lines — e.g. an
Einstein-solid-style hydride evaluated without a translational component.
Decks with twt > 0 never enter this branch (the lines are convolved
through the translational path instead), which is presumably why the
standard evaluation suite never exposed it.

Reproduction sketch: a minimal one-temperature deck with a small
continuous spectrum (tbeta = 0.5), twt = 0, and two oscillators
(nd = 2, e.g. 0.137 eV and 0.250 eV with weights 0.3/0.2) on a beta grid
reaching past both lines. The resulting S(α,β) carries a delta spike only
at the first oscillator's beta; the second line (and the
oscillator-combination lines) are absent. Note the iprint >= 2
"norm check" printout still reports the full line-weight sum, because it is
computed before this block — so the lost weight is not visible in the
diagnostic output.

Suggested fix: use a separate flag for the inner grid search, e.g.

                     jdone=0
                     j=0
                     do while (j.lt.nbeta.and.jdone.eq.0)
                        ...
                           jdone=1
                        ...
                     enddo

so the outer loop runs over all n lines. (The early exit on
dwf.lt.vsmall is unaffected and can stay on the outer flag.)

We found this by cross-checking against an independent reimplementation
that convolves every in-range line: on the two-oscillator deck above it
produces distinct delta spikes at both oscillator betas, while NJOY2016.78
produces only the first.