SQL Layer Hardening: Implement sql_buffer_t and refactor bulk queries

[somethingwithproof]

Summary

I’m planning a follow-up hardening pass for the Spine SQL layer. Currently, we build our bulk queries (like the main data push in util.c) using dozens of manual sqlp += sprintf(...) calls.

Work

This approach is fragile for two reasons:

1. It relies on manual pointer arithmetic and buffer tracking at every single step, which is a prime candidate for off-by-one errors or silent overflows if a field grows.
2. It’s a maintenance headache—every time a field is added to the schema, we have to manually sync several blocks of repetitive sprintf logic.

The Plan:

1. Implement a small sql_buffer_t helper that handles bounded snprintf and pointer advancement automatically (eliminating the manual += math).
2. Refactor poller_push_data_to_main to use this helper. This will likely shrink util.c by ~100 lines and make the SQL structure much easier to audit and maintain.

Goal is to make the data path 100% 'safe-by-construction' and easier for us to add fields in the future without the boilerplate noise.

Acceptance Criteria

• Work items are implemented and validated.
• Changes preserve behavior unless explicitly intended.

[somethingwithproof]

Updated PR #467 with critical data integrity fixes for NULL columns and db_escape hardening.

[somethingwithproof]

Implemented in PR #511. Ready for review.