fix(studio): preserve caption text when parsing transcripts

[AksharP5]

Caption editing can miss an entire transcript when a word contains ];. The old normalization also changes literal text such as ,] and rejects JavaScript escapes.

Use Acorn to find the array boundary and decode static data without executing the source. Text, word IDs, timing, and existing JSON/single-quoted syntax are preserved. Executable expressions and malformed input still return an empty transcript.

Fixes #3852.

Verification

• Reproduced five regression failures on upstream before the fix. All 92 caption tests now pass, including malformed input and non-execution checks.
• Real-browser check of parseCaptionComposition with an iframe DOM preserved x ]; y, literal ,] and ,}, hex escapes, escaped quotes, IDs, and timing. An executable initializer was rejected with zero calls.
• Studio typecheck and build, full workspace build, oxlint, oxfmt, and the fallow diff gate pass.

Four files changed. No visual styling or layout changes.