Decide tech stack

[pphulara]

What this PR does

Fills in the Tech Stack and System Design sections of docs/architecture.md, which were both marked TODO.

Stack decided

Layer	Choice
Backend	Laravel (PHP)
Admin panel (TPO)	Filament v5 + Livewire
User-facing pages (alumni, students)	Svelte 5 + Inertia.js
Real-time (chat, notifications)	Laravel Reverb
Database	PostgreSQL + pgvector
Multi-tenancy	stancl/tenancy
AI (future, not V1)	Laravel Prism

Why this stack

Six dimensions were researched and weighed:

1. Low-end devices + slow networks — primary constraint. TPOs at Tier 2/3 colleges are on weak hardware and slow connections. Filament's server-rendered output and Svelte's 15–40 KB compiled JS both work. No heavy SPA shipped to the client.
2. Admin/CRUD tooling — V1 is an admin tool first. Filament v5 is the strongest batteries-included admin panel in the PHP/JS ecosystem. Theming (shadcn, Aura UI) brings it to 2026 SaaS aesthetics.
3. Multi-tenancy — stancl/tenancy supports shared-DB and DB-per-tenant from the same package. Start with shared DB for V1 (KEC alone), graduate to DB-per-college when needed. No rewrite.
4. Hosting cost — PHP runs on shared hosting in India from ~₹400/month. Cheapest entry point for a college-budget product.
5. AI-readiness — Laravel Prism covers LLM API calls. pgvector on PostgreSQL handles embeddings. No separate Python service needed for the alumni-matching / NL-search roadmap.
6. Scalability — same stack runs multi-tenant SaaS at 1,000+ tenants, real-time social platforms, and LinkedIn-scale activity feeds in production.

Internal precedent: ColoredCow shipped Plio (multi-tenant, low-bandwidth, ed-tech) on Django + Vue. Laravel's multi-tenancy tooling (stancl/tenancy) is more flexible than django-tenants for the "start shared, graduate to DB-per-tenant" path this product needs.

What this PR does NOT decide

• Data model schema
• Final V1 feature scope
• Hosting platform and CI/CD setup