Correct q and COCOS statements in sign conventions reference

[krystophny]

Fixes #267. Follows up on @logan-nc's review comment on #253: the sign conventions reference merged in #266 claims q is read directly from the EFIT g-file without sign manipulation. For g-file input, q is recomputed by field-line integration in direct_run (equil/direct.f) and always comes out positive; the file's q profile is read but unused.

Changes to docs/sign_conventions.rst:

• Safety Factor section: q is recomputed, not read; always positive for g-file input; inverse equilibria keep the input q profile including its sign; negative newq0 remains the one override.
• "Why Positive m Is Always Resonant": q > 0 is now stated as guaranteed for direct equilibria rather than "standard tokamak".
• COCOS Compatibility: sign choices in the g-file are normalized away on read (psi flip + ABS(F) + recomputed q), so only the per-radian flux units (Wb/rad) are required, not one specific sign convention.
• Quick Reference table and Source Code References updated to match.
• Two LaTeX-style em-dashes in the helicity section replaced.

Same correction applied to the q bullet in CLAUDE.md.

Verification

Documentation claims tested against develop (375ee8f) built with gfortran 16.1.1.

Failing before (doc claim "read directly from the EFIT g-file without sign manipulation" is falsified by the code): DCON run on the DIII-D ideal example g-file versus a copy with all signed quantities negated (psirz, simag, sibry, cpasma, ffprim, pprime, qpsi; the file's q profile becomes -1.06 to -8.03):

baseline: q0 =  1.062E+00, qmin =  1.062E+00, qmax =  5.320E+00, q95 =  4.228E+00
flipped:  q0 =  1.062E+00, qmin =  1.062E+00, qmax =  5.320E+00, q95 =  4.228E+00

$ cmp baseline/euler.bin flipped/euler.bin && echo IDENTICAL
IDENTICAL
$ cmp baseline/crit.bin flipped/crit.bin && echo IDENTICAL
IDENTICAL
$ diff baseline/dcon.out flipped/dcon.out
<  Total cpu time =  4.397E+00 seconds
<  Total wall time =  4.481E+00 seconds
>  Total cpu time =  4.468E+00 seconds
>  Total wall time =  4.525E+00 seconds

The g-file q sign has no effect and the recomputed q is positive in both runs, matching the corrected text. The unmodified example file already has fpol < 0, so the ABS(F) branch is exercised in the baseline; the flipped file has psio < 0 and exercises the psi flip branch.

Passing after: the revised rst parses cleanly (docutils; the pre-existing Sphinx-only :doc: role is unchanged).

[logan-nc]

Superseded by #269, which carries these commits plus review corrections: the inverse-equilibrium q statement here was inaccurate (the main inverse path inverse_run recomputes q like direct_run rather than keeping the input file's q), and #269 also adds positive-q enforcement (guards newq0 < 0 and a chease4 q check) and removes the dead newq0==-1 sign-flip code. Opened on a PrincetonUniversity/GPEC branch because the maintainers can't push to this fork. This PR can be closed in favor of #269.

[krystophny]

Great, thanks Nik!