fix(data): resolve 364 smartphone brand-prefix conflicts

[Seungpyo1007]

Summary

Follow-up to #329. That PR merged the 50 brand-prefix pairs whose core specs were
already identical and left 717 pairs that disagreed on a core field
(soc / release_date / ram_gb / battery_mah / weight_g / os) documented
but unresolved. This PR resolves the largest, highest-confidence slice.

How the 717 break down

Every pair is a bare slug (dx650, from the Kaggle gsmarena-mobile-devices dump,
carrying a raw_gsmarena block) vs a brand-prefixed twin (acer-dx650). Conflict
fields across the 717: soc 572, release_date 498, weight_g 159, battery_mah 83,
ram_gb 62, os 14.

The soc conflict is almost never a real spec difference — both slugs resolve to a
real soc record, and in 414 pairs they are the same physical chip under a
different slug spelling (qualcomm-msm7227-snapdragon-s1 vs qualcomm-msm7227,
qualcomm-msm8260a-snapdragon-s4-plus vs qualcomm-snapdragon-s4-msm8260a). The
remaining conflicts on those pairs are the fabricated bulk-import placeholders
(weight_g=150.0, battery_mah=1500, generic YYYY-01-01 date) that #329 flagged.

What this PR changes: 364 merges

The 364 same-chip pairs sourced Kaggle-GSMArena (bare) vs phonedb.net (prefixed).
Keeper = the GSMArena record (real specs, richer, has the raw_gsmarena provenance
block). For each:

• soc: kept the keeper's slug — confirmed FK-valid for all 364, same physical chip.
• weight_g / battery_mah: kept the keeper's GSMArena value; the phonedb twin's
  150.0 g / 1500 mAh are the stamped placeholders. In all 364 the keeper already
  held the real value, so no scalar was rewritten — the placeholder only ever
  lived on the deleted twin. Spot-verified: Acer DX650 = 133 g (GSMArena,
  imei.org, softpedia) not phonedb's 150.0; Acer CloudMobile S500 = 122 g.
• name: upgraded to the brand-inclusive form (Acer DX650 over DX650).
• source_urls: unioned, so the phonedb.net reference is retained.
• verified: left unchanged — deliberately not flipped to true, because the
  phonedb side's scalars proved unreliable, so it isn't a trustworthy source.
• release_date: left as the keeper's value and not counted as resolved.
  phonedb's specific dates are unreliable (it lists Acer DX650 as 2009-03; GSMArena
  and imei.org say 2009-06), so importing them would be guessing. Reported as
  still-unresolved rather than changed.
• The redundant phonedb twin file is deleted.

No value was invented.

Counts (this PR)

• resolved-and-merged: 364 (same-chip Kaggle-vs-phonedb duplicates)
• confirmed-different-devices-not-duplicates: 0 in this PR (the ~158 different-chip
  pairs, which include the genuine variants, are being triaged separately)
• still-unresolved-no-source (carried to follow-up): 353 — 37 same-chip pairs
  sourced vs vendor/Wikipedia pages, 157 needing a per-record ram/weight/battery
  source, 158 different-chip pairs (some genuinely different devices to un-pair),
  plus the release_date field on the 364 merged here.

Validation

• python -m app.validate → passed.
• integrity_check.py <data> --strict → no hard anomalies (0 duplicate slugs,
  0 slug/file mismatches, 0 verified-without-source).
• Post-merge: 92,994 smartphone files (−364), zero duplicate slugs.

Dump

Last commit refreshes site/public/v1 from the merged data (python -m app.dump,
no re-scrape): removes the 364 deleted slugs and reconciles score percentiles.

Closes #296