feat: spell out dev acronyms espeak mispronounces in normalize

[StuBehan]

Summary

espeak mispronounces several common dev acronyms — it reads them as a word instead of spelling them out (CLI→"kligh", CI→"sigh", IDE→"eyed", AWS→"awz", URI→"yuri", IAM→"eeyam") and mangles CI/CD into "sigh slash C-D". This adds a small, curated pronunciation map applied by default in normalize, so read-aloud output says "C-L-I", "A-W-S", "C-I-C-D", etc.

Changes

• _DEV_PRONUNCIATIONS map in stackvox/text.py, applied by default via a new dev_terms=True parameter on normalize_for_speech. Whole-word and case-insensitive (reusing apply_pronunciations), so lowercase cli is fixed too. Caller-supplied --pronunciations still override the defaults.
• --no-dev-terms flag on the normalize / say / speak subcommands to opt out.
• Curated to only the terms espeak gets wrong. Acronyms it already voices correctly — API, URL, JSON ("jason"), YAML, HTTP, SDK, nginx ("engine-x"), CRUD, REST — are deliberately left untouched. Plus a few said-as-a-word fixes: SaaS→"sass", PaaS→"pass", TUI, postgresql→"postgres", kubectl→"kube control".
• Doc reconciliation (docs/speech-normalization.md): clarifies that generic pronunciation corrections live in core, while domain/org dictionaries (StackOne, Redis, …) stay per-consumer.

No change to correctly-voiced text; opt-out preserves the old behaviour exactly.

Testing

• pytest — 181 passed; ruff check . clean.
• Pronunciations were chosen empirically via Kokoro().tokenizer.phonemize(term, "en-us") (no audio needed): confirmed the six target acronyms read as words before and spell out after, and that API/URL/JSON/HTTP/SDK were already correct (hence excluded).
• End-to-end:

  $ stackvox normalize "Use the CLI to deploy to AWS. The CI/CD pipeline parses the URI for IAM."
  Use the C L I to deploy to A W S. The C I C D pipeline parses the U R I for I A M.
  $ stackvox normalize --no-dev-terms "Use the CLI to deploy to AWS."
  Use the CLI to deploy to AWS.
  

Related issues

None.