itk::StringTo* helpers accept trailing content after a valid numeric prefix

[hjmjohnson]

The itk::StringTo* helpers in Modules/Core/Common/src/itkStringConvert.cxx accept a valid numeric prefix followed by arbitrary trailing content. itk::StringToDouble("1.25junk", ctx) returns 1.25 rather than throwing, because the validity check is "did the conversion start" (end == begin) rather than "did the conversion consume the input".

Every IO call site migrated to these helpers in #6089 inherits that leniency, so a malformed header field or DICOM Decimal String can silently yield a plausible-looking value instead of a parse error.

Current behavior

Measured against ITKCommon (macOS arm64), alongside the std::stod these helpers replaced:

Input	itk::StringToDouble	std::stod
"1.25junk"	1.25	1.25
"1.25 "	1.25	1.25
" 1.25"	1.25	1.25
"1.25\2.5"	1.25	1.25
"junk"	throws	throws
"1.25"	1.25	1.25

The behavior is identical to std::stod, so it long predates the locale work in #6695; it is inherited, not introduced. strtod's endptr contract distinguishes "nothing consumed" from "something consumed" and leaves "partially consumed" to the caller, and std::stod chose the lenient reading.

Why this is not a one-line fix

A naive *end != '\0' rejection would break valid input:

• DICOM DS explicitly permits leading and trailing spaces, so "1.25 " must keep parsing. Any strict check has to skip trailing whitespace before rejecting.
• itk::StringToDouble / StringToFloat are now the shared funnel for many readers (NRRD, DICOM, and others migrated in BUG: Wrap std::sto* exceptions in IO with itk::ExceptionObject (#3213) #6089). Tightening validation is an input-compatibility change: files that load today could start failing, which is a deliberate decision rather than a drive-by fix.

Proposal

1. Reject trailing content that is not whitespace, for the whole itk::StringTo* family (integer helpers included — std::stoll/std::stoull are lenient the same way).
2. Keep leading and trailing whitespace valid.
3. Add GTest coverage for the prefix cases: "1.25junk", "1.25\2.5", "1 2", "0x10", and the whitespace cases that must still pass.
4. Survey call sites for readers that legitimately hand these helpers a substring with a delimiter still attached, and fix those to split first rather than relying on the lenient parse.

Provenance

Raised as a P1 by an automated reviewer on #6695 (discussion_r3646959273), attributed there to that PR's strtod_l change. Verification against the merge base showed the behavior unchanged from std::stod, so it was kept out of #6695's scope and tracked here instead.