feat(ductile-fracture): load finalized plots, SEM datasets and dictionary review #233

[pierreleripoll]

Ductile Fracture — finalized data load (#233)

Loads the finalized ductile-fracture dataset delivered by @andskia (replacing the
single-specimen placeholder), applies his data-dictionary review, and enables SEM
for the specimens that now have imagery. Stacks on the material/ nesting (#37,
already merged).

Contents

• Plots — all 7 series for 102 specimens:
  • force_elongation (102), force_radial (85 — 17 specimens have no
    experimental radial data; simulation-only for those), and the five
    *_simulation series (102 each: force-elongation, force-radial, equivalent
    plastic strain, lode angle, triaxiality).
• SEM — 36 specimens added under sem/<id>/ (stored as Git-LFS, per the
  existing *.tif filter):
  • 23 single-surface CNT (5 locations), 12 GP (3 locations: east/middle/west),
    and 12 Ozden et al. CNT coupons imaged on both fracture surfaces
    (_2 suffix → surface 2).
• Data dictionary — applied @andskia's 18 reviewed descriptions; resolves
  most (?) markers (piece, crack_opening_direction, exc_at_fract,
  step_fract, CNT diameter fields, …).
• has_sem — set to yes for the 36 SEM specimens, no otherwise.

Validation

Ran the full uploader (all-ductile-fractures) against a clean local Postgres —
zero errors. Row counts: 102 parent specimens, has_sem=true = 36,
force_radial covers 85 distinct specimens, all 7 series tables populated.

Reviewer / @andskia follow-ups (non-blocking)

• Pending symbols — swdfm_* and kk_* rows have descriptions but their
  symbol column is still empty (Andskia's @add symbol@ notes).
• Model attribution — kk_* descriptions were changed Kanvinde-Kiran →
  Kiran-Khandelwal per the review; please confirm.
• elongation_fr still carries a (?) (not covered by the review).
• Dropped file — a lone extra_017.tif in
  Skiadopoulos_..._CNT_EXT_M_C3_R10_0 (location "extra", magnification 17) was
  excluded as it fits no naming/magnification convention. Confirm if it was
  meaningful.

Notes

• SEM CDN assets are not in this PR — the WebP/manifest.json for the 36
  SEM specimens are produced + uploaded separately via make precompute-sem
  (EPFL CDN). Until that runs, the 36 SEM dialogs will 404.
• Squash-merge (repo convention — post-merge CI re-validates HEAD~1..HEAD).