fix(chat): only leave agent mode when the model really has no tool support

[MummIndia]

Summary

Both checks in static/js/chat.js match a bare tool or auto anywhere in the error
text, which catches a great deal that has nothing to do with tool support: a tool
timing out, an MCP server dropping, any message containing "automatic".

At the first site the consequences are not cosmetic. The real error is replaced with
"This model doesn't support agent tools", so the actual cause is lost. The UI leaves
agent mode. And the choice is written to localStorage, so every later message in
that conversation also runs without tools, with nothing on screen explaining why. It
presents as "agent mode works in a new chat but not in this one", which sends you
looking at the model, the session or the endpoint rather than at a substring match in
the error path.

The second site is milder — it appends a hint rather than switching — but appending
"this model may not support tools" to an unrelated failure still points the user at
the wrong fix.

This matches the provider's actual wording instead. Ollama returns <model> does not support tools; the pattern also covers "doesn't support tools" and "tools are not
supported". The original error text is kept in both branches so a genuine failure
stays diagnosable.

Target branch

• This PR targets dev, not main.

Linked Issue

Fixes #5766

Type of Change

• Bug fix (non-breaking — fixes a confirmed issue)

Checklist

• I searched open issues and open PRs — this is not a duplicate.
• This PR targets dev
• My changes are limited to the scope described above — no unrelated refactors or whitespace changes mixed in.
• I actually ran the app (docker compose up) and verified the change works end-to-end.

How to Test

1. In agent mode with a model that does support tools, trigger a failure whose message
   contains tool or auto — killing an MCP server mid-turn, or letting a tool time
   out, both work.

2. Before the change — the message shown is "This model doesn't support agent
   tools", the mode toggle flips to Chat, and localStorage keeps it there. Sending a
   second message in the same conversation runs without tools.

3. After the change — the real error text is displayed and agent mode stays on.

4. Regression check on the strings that used to misfire — none of these demote any
   more, while the three real phrasings still do:

   const re = /does\s*n[o']?t\s+support\s+tools?|tools?\s+(?:are\s+)?not\s+supported/i;
   
   re.test('registry.ollama.ai/library/llama3.2 does not support tools'); // true
   re.test("this model doesn't support tools");                           // true
   re.test('Tools are not supported by this endpoint');                   // true
   
   re.test('Error 500: tool execution timed out');        // false
   re.test('MCP server not connected: builtin_browser');  // false
   re.test('automatic retry failed');                     // false
   re.test('tool_choice auto is invalid');                // false
   re.test('Error 502: peer closed connection');          // false

Visual / UI changes — REQUIRED if you touched anything that renders

This touches static/js/chat.js, so I want to be explicit rather than tick "none".

What changes on screen: an error message string, and the state of the mode
toggle in the cases where it previously flipped by mistake. No styling, layout,
spacing, colour, font, icon, HTML structure or CSS is touched — the diff is a regular
expression and the two conditions that use it.

I have not attached a screenshot because producing one means provoking an artificial
backend failure, and the resulting image would show one arbitrary error string in a
message bubble — it would not illustrate the defect, which is which errors are
matched. The regression table above is a more faithful account of the change.

Happy to attach a before/after capture if you would rather have one — say the word and
I will provoke the failure and record it.