fix: improve rich text and analysis loading (#1092)

* fix(api): type rich text limit failures

Return expected rich-text size failures as typed success:false API responses instead of relying on request-schema 400s or frontend Axios error parsing.

Changes:

- add shared rich-text validation reasons for plain-text and raw-HTML limits

- validate sanitized conversation descriptions and statements in service code

- return typed failures from create/update conversation and create opinion

- update frontend submission paths to branch on typed response reasons

- regenerate OpenAPI clients for agora and load-testing

- rename API image scripts from production-specific names to generic tag/push commands

Deploy: agora, api

* refactor(workers): rename analysis shared package

Rename the Python shared package used by math-updater and description retry workers from the generic python-worker-shared/agora-worker-shared naming to analysis-worker-shared/agora-analysis-worker-shared.

Changes:

- move the package directory and import module to analysis-scoped names

- update math-updater, ai-description-retry-worker, and description-translation-retry-worker imports, Dockerfiles, pyproject files, and uv locks

- update Python artifact generation paths for math-updater generated models and shared types

- make AGENTS.md wording agent-agnostic

- move legacy root script infrastructure assets under scripts/infra

Deploy: math-updater, ai-description-retry-worker, description-translation-retry-worker

* docs: prune obsolete project artifacts

Remove old strategy notes, deprecated architecture sketches, generated documentation site assets, and stale production public key files.

Keep the embed guide by moving it from doc/old/embed.md to doc/embed.md.

Deploy: none

* fix(agora): show loader when switching analysis variants

Prevent the analysis tab from rendering content from a previous group-count variant while an uncached selected variant is still loading. Initial analysis rendering now waits for checkpoint data too, so the checkpoint timeline does not appear in a partial state.

Keep same-key refresh behavior intact for SSE and pull-to-refresh by removing cross-key placeholder data instead of hiding refetches with existing data. Analysis frame-section fetches now use explicit fail-fast retry behavior to avoid hidden repeated expensive requests during foreground loads.

Add a dev component-testing route that reuses AnalysisPage with mock analysis data and a cache reset button for testing slow uncached variant transitions.

Deploy: agora

* refactor(workers): rename shared analysis package

Rename the shared Python analysis package directory to services/shared-analysis-worker so it matches the repository's shared-* package naming convention. Update editable dependency names, Docker install paths, codegen outputs, docs, uv locks, and generated model metadata accordingly.

Rename the schema codegen selector from math-updater to shared-analysis-worker to reflect that generated SQLAlchemy models are shared by math-updater and the description retry workers. Runtime math-updater dev/log targets remain unchanged because they still run the math-updater process.

Also remove unused rich-text validation schemas and private unused dev fixture helpers, and simplify the Python shared-types generator test so it reads real shared source files instead of duplicating constants inline.

Deploy: math-updater, ai-description-retry-worker, description-translation-retry-worker

* fix(agora): skip empty seed delete dialog

Remove empty seed statements immediately when the delete button is clicked instead of opening the destructive confirmation drawer. Non-empty rich text still opens the confirmation dialog, using the same counted text logic as editor validation to treat empty rich-text markup as empty content.

Deploy: agora

Author: nicobao

AGENTS.md

@@ -1,12 +1,12 @@
-# CLAUDE.md
+# AGENTS.md
 
-This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+This file provides guidance to AI coding agents when working with code in this repository.
 
 ## External Repository References
 
 When the user mentions needing to understand an external library or repository, **first check if it's cloned locally** in the `external/` directory at the repository root before fetching from GitHub or using web search.
 
-**Important**: The repository root is where CLAUDE.md is located. Claude's working directory may be a subdirectory (e.g., `services/agora/`), so use paths relative to the repository root or use `../../external/` to navigate up from subdirectories.
+**Important**: The repository root is where AGENTS.md is located. An agent's working directory may be a subdirectory (e.g., `services/agora/`), so use paths relative to the repository root or use `../../external/` to navigate up from subdirectories.
 
 **Pattern**: External repos are sometimes cloned into `external/` at repository root for analysis. This directory is gitignored.