fix(instr): case-insensitive match and fix wrapped-substr clobber (#3650)

[chiliec]

What

Closes #3650.

INSTR had two independent defects in sql/expression/function/substring.go:

1. Case-sensitive for nonbinary strings. findSubsequence compared
   runes exactly, so INSTR('xyza','A') returned 0. Per the MySQL 8.4
   manual, INSTR "is multibyte safe, and is case-sensitive only if at
   least one argument is a binary string" — it should return 4. This
   mirrors the existing case-insensitive behavior of LOCATE
   (locate.go already lower-cases both sides).

2. Wrapped substring clobbered the haystack. In Instr.Eval, the
   sql.StringWrapper case of the substring argument assigned the
   unwrapped value to text (the haystack) instead of subtext (the
   needle). A wrapped substring (e.g. Dolt's out-of-band TextStorage
   values) therefore left subtext empty and overwrote text, so
   findSubsequence ran with an empty needle and INSTR returned 1
   unconditionally.

Fix

• findSubsequence now folds case per rune via a small runeEqualFold
  helper (a == b || unicode.ToLower(a) == unicode.ToLower(b)).
• The sql.StringWrapper branch of the substring argument now assigns to
  subtext.

+10/-2 in substring.go, plus one new test file.

Tests

Added TestInstrIssue3650 covering both defects (case-insensitive
needle/haystack/mixed + no-match, and wrapped-substring match/no-match/
case-insensitive).

Validation (real results)

Toolchain: Go 1.26.2, -tags gms_pure_go (pure-Go regex engine, no cgo).

• RED→GREEN proven. With the new test in place and the source fix
  reverted (git checkout on substring.go), all 6 sub-tests fail
  (case-insensitive cases return 0, wrapped-substr cases return 1).
  Restoring the fix turns them green:

  # reverted source, test kept:
  --- FAIL: TestInstrIssue3650/case-insensitive_needle
  --- FAIL: TestInstrIssue3650/wrapped_substr_match
  ... (6 sub-tests fail)
  # with fix:
  ok  github.com/dolthub/go-mysql-server/sql/expression/function
  

• Existing TestInstr still passes.
• gofmt -l clean on both files.

The full ./sql/expression/function/ package suite passes except for
TestRegexpReplaceWithFlags/{sensitive_and_multiline_flags,all_flags},
which fail identically on a clean main checkout under the
gms_pure_go tag (a CRLF difference in the pure-Go regex engine) — they
are outside this diff and unrelated to this change.

First-time contributor here — happy to add an enginetest/queries
end-to-end case or adjust anything if you'd prefer.