fix(beta/network): resolve namespaced and enumerate unknown delegate targets

[genisis0x]

Why are these changes needed?

A delegate target that doesn't resolve sends the run into a retry loop: the model gets a bare Error: target ... not found, re-guesses, and in multi-agent pipelines stalls out (reported in #3022, an extension of #2991 / #2998). The delegate call itself fails fast (~0.6ms in the telemetry on #3022) — the timeout is the model never converging on a real target.

This addresses both flavours of that loop:

1. Namespaced-but-valid targets. LLMs occasionally echo OpenAI's functions. tool namespace into the target string — e.g. functions.evaluator instead of evaluator. The literal name misses the hub registry, so the consult would fail even though the peer is real. Resolution now tries candidate names most-literal first: the name as given, then — only when it carries the functions. prefix — the unprefixed name. A valid peer behind a namespaced target resolves and the consult proceeds. The channel is opened with the resolved name.

2. Fully invented targets. When no strip recovers the name (e.g. PPTDialogScript), the model has no signal about which peers are real, so it keeps re-guessing. The not-found error now enumerates the live directory — Error: target 'PPTDialogScript' not found. Available targets: [evaluator, translator] — so the next turn can pick a real one. The lookup is best-effort: a directory failure falls back to the bare not-found error, and the caller is excluded since self-delegation is rejected anyway.

Genuinely unknown targets still fail fast (now with the available-targets hint), and self-delegation still short-circuits as added in #2998.

Related issue number

Closes #3022

Checks

• I've included any doc changes needed for https://docs.ag2.ai/.
• I've added tests (if relevant) corresponding to the changes introduced in this PR.
• I've made sure all auto checks have passed.

AI assistance

• I understand the changes in this PR and can explain them in my own words.
• I have verified that the PR description accurately reflects the actual diff.
• If AI assistance was used, I reviewed, tested, and validated the generated code/text before submitting.

[codecov]

Codecov Report

❌ Patch coverage is 88.00000% with 3 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
ag2/network/client/tools/delegate.py	88.00%	2 Missing and 1 partial ⚠️

Files with missing lines	Coverage Δ
ag2/network/client/tools/delegate.py	81.08% <88.00%> (+4.15%)	⬆️

... and 4 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[genisis0x]

@Lancetnik whenever the network-area queue allows, no rush — this one's green and folds the durable half of #3022: the enumerate-unknown-delegate-targets change, so a not-found delegate now returns an error listing the live registered peers instead of letting the model keep hallucinating a target until it times out. Happy to rebase if it's drifted.