feat: add live context token counter after each orchestrator loop

[carTloyal123]

Summary

Adds an inline debug-style context usage line after each LLM round-trip in the orchestrator loop, so users can monitor context growth in real time during multi-tool-call turns.

What it does

After each iteration of the orchestrator's while !should_yield loop (i.e., after the LLM response + tool results are folded into the context), a TitleFormat::debug message is emitted showing:

• Current token count (actual or approximate with ~ prefix)
• Compaction threshold (if configured, shown as / 900.0k)

Example output:

● [14:32:05] Context 45.2k / 900.0k
● [14:32:12] Context ~82.1k / 900.0k

Changes

• crates/forge_app/src/orch.rs: Added emit_context_usage() helper and humanize() formatter. Emits context line after each loop iteration at the point where context is fully updated.
• crates/forge_domain/src/agent.rs: Added unit test test_compaction_threshold_claude_opus_4_6_with_1m_context_window verifying the compaction threshold calculation for 1M-context models.

Testing

• New unit test passes: cargo test -p forge_domain test_compaction_threshold_claude_opus_4_6
• cargo check passes across workspace
• Manually verified inline output during multi-tool-call sessions

Co-Authored-By: ForgeCode noreply@forgecode.dev

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}