This closes #2217: Preserve editAs positioning when copying pictures with twoCellAnchor

[blankInPajamas]

Fixes #2217

Problem

Copying a picture via AddPictureFromBytes produced a visibly stretched/
distorted copy when the source picture was anchored with a twoCellAnchor
element carrying an explicit editAs="oneCell" attribute. This is a valid,
spec-compliant anchor pattern commonly produced by WPS Office (and other
third-party tools), distinct from excelize's own writer, which always emits
a true oneCellAnchor element for size-locked pictures.

Root cause

Two issues compounded to cause this:

1. extractPictureFromAnchor and extractPictureFromDecodeAnchor only
   inferred Format.Positioning = "oneCell" when the anchor had no <to>
   element at all. They never checked the anchor's own editAs attribute,
   so a twoCellAnchor with editAs="oneCell" (which does have a <to>)
   was never recognized as size-locked.

2. For anchors that fall through to the fallback decode path
   (extractDecodeCellAnchor, used when a drawing part mixes chart/
   graphicFrame anchors together with picture anchors), the re-parse of the
   anchor's inner content did not preserve the parent anchor's editAs
   attribute, so it was lost even when correctly read initially.

The end result: Format.Positioning came back empty for these pictures, so
AddPictureFromBytes wrote a plain twoCellAnchor with no editAs on
copy — making the copied picture resize with its cells instead of keeping
its original locked size.

Fix

• Recognize editAs directly when reading a picture's anchor, falling back
  to the existing "oneCell when no to" inference only when editAs is
  absent.
• Preserve the parent anchor's editAs when the decode-fallback path
  re-parses anchor content.

Testing

• Added a new case to TestGetPicture covering a twoCellAnchor with
  editAs="oneCell" and a to element present, asserting GetPictures
  returns Format.Positioning == "oneCell".
• Verified against a real .xlsx file produced by WPS Office: before the
  fix, copying the picture dropped editAs="oneCell" from the resulting
  drawing*.xml (becoming a plain twoCellAnchor, which visibly stretched
  the picture); after the fix, the copy correctly writes a proper
  size-locked oneCellAnchor, matching the source picture's appearance.
• Ran the full test suite (go test ./...) with no regressions.
• Proceeded to run (gofmt -l) . and (go vet ./...) and found no errors