Stop forcing *increments() columns to be the table primary key

[eaguad1337]

Migrated from MasoniteFramework/orm#950 (original author: @circulon) as part of the move to the masonitedev organization. Ported, evaluated, and hardened for 3.x.

Closes #4

What this does

increments(), tiny_increments() and big_increments() no longer implicitly mark the column as the table's primary key, aligning them with every other datatype usable as a primary key. This unblocks tables that pair a non-integer primary key (e.g. uuid) with an auto-numbered column — previously that raised QueryException: table has more than one primary key:

with self.schema.create("invoice") as table:
    table.uuid("id").primary()
    table.big_increments("reference")   # now works
    ...

• .primary() on an *increments() column compiles the PRIMARY KEY inline through dedicated *_increments_primary type mappings.
• id() keeps the previous convenience behavior (auto-incrementing big-integer primary key).
• The internal migrations table schema declares .primary() explicitly.
• Schema now honors explicitly provided connection_details instead of always reloading them from the config file (same pattern #948 introduced for the QueryBuilder).

⚠️ BREAKING

Migrations relying on *increments() implying the primary key must add .primary():

table.increments("id")            # 3.0.x: primary key — now: plain auto-increment (raises on SQLite)
table.increments("id").primary()  # previous behavior
table.id()                        # unchanged convenience

Platform-SQL fixes added on top of the original PR

The original mappings compiled to invalid SQL on two platforms:

Platform	Original PR	This PR	Why
SQLite	BIGINT PRIMARY KEY AUTOINCREMENT	INTEGER PRIMARY KEY AUTOINCREMENT	AUTOINCREMENT is only valid on INTEGER PRIMARY KEY; the tiny/big variants were syntax errors. The rowid alias is 64-bit, so big_increments loses no range.
MSSQL	INT PRIMARY KEY IDENTITY	INT IDENTITY NOT NULL PRIMARY KEY	T-SQL grammar puts IDENTITY before column constraints; PRIMARY KEY is now emitted via the constraint slot after nullability.
MySQL (non-primary)	INT UNSIGNED AUTO_INCREMENT	INT UNSIGNED AUTO_INCREMENT UNIQUE	MySQL requires an AUTO_INCREMENT column to be defined as a key (error 1075).
Postgres (non-primary)	SERIAL	SERIAL UNIQUE	Keeps the UNIQUE the old SERIAL UNIQUE mapping carried; consistent with MySQL.

SQLite still raises a descriptive QueryException for non-primary *increments() (SQLite cannot autoincrement a non-PK column).

Tests

• Split the SQLite unsupported-types test so each variant actually asserts (blocks 2 and 3 were dead code inside the first assertRaises).
• Added compile assertions for id() and big_increments().primary() on SQLite plus an executed id() table creation against the real test database, so invalid AUTOINCREMENT DDL can't slip through string-only tests again.
• Changelog entry moved to an Unreleased section (3.0.0 already shipped without this change).
• Full suite: 1037 passed, 1 pre-existing failure (requires a live Postgres server — fails identically on 3.x).