fix: avoid false italic conversion in Telegram reasoning

[cpfiffer]

Summary

• tighten Telegram reasoning markdown-to-HTML conversion for *italic* and **bold**
• avoid converting math/operator asterisks like 2 * 3 * 4 or 2*3*4
• add regression tests for spaced and compact math expressions

Why

PR #523 added markdown conversion for Telegram thinking blocks, but the broad italic regex also matched non-markdown uses of asterisks.

Testing

• npm test -- src/core/display.test.ts
• npm run test:run

[cpfiffer]

Quick context on what happened:

• I reviewed the original Telegram reasoning change from fix: convert markdown bold/italic to HTML in Telegram thinking blocks #523 and found a follow-up bug: the italic conversion regex was too broad and treated math/operator asterisks as markdown emphasis.
• Repro example: 2 * 3 * 4 was being rendered as 2 <i> 3 </i> 4 in Telegram HTML mode.
• fix: convert markdown bold/italic to HTML in Telegram thinking blocks #523 is already merged, so this is a follow-up fix PR.

What this PR changes:

• Tightens the markdown-to-HTML emphasis regex in formatReasoningDisplay so it only converts likely markdown delimiters.
• Avoids converting escaped/in-word/math uses of *.
• Adds regression tests for both 2 * 3 * 4 and 2*3*4.

Validation:

• npm test -- src/core/display.test.ts
• npm run test:run

Both pass.

[cpfiffer]

Leaving this as a follow-up note so we can decide intent later.

I think a sensible default is to treat *...* / **...** as formatting only when delimiters are clearly markdown (not in-word operators), and to apply that consistently across languages.

Right now the fix covers ASCII math cases (2 * 3 * 4, 2*3*4), but there may still be Unicode in-word edge cases due to ASCII-oriented boundaries.

When we revisit, I suggest we confirm intent and add two focused tests:

• привет*мир* stays literal
• привет *мир* formats мир as italic