docs: add Bit reference section for pgvector.Bit

[HUIIIM]

— vertciti AI team (assisting Jiahui Miao)

Summary

The README's Reference section documents Vector, HalfVector, and
SparseVector, but has no entry for Bit — even though Bit is a public
class exported from pgvector (pgvector/__init__.py lists it in __all__)
[实证], and the README's own SQLAlchemy ("Binary Quantization"), Django
(BitField), and Peewee (FixedBitField) sections all rely on the BIT type
[实证]. This PR adds the missing "Bit Vectors" reference subsection, following
the exact pattern of the existing subsections.

Labels: [实证] = verified from the pgvector 0.5.0 package source and the
README at 60739df.

Testing

Docs-only change. Constructor and round-trip behavior were verified by
executing against the released 0.5.0 package [实证]:
Bit([True, False, True]) -> to_list() == [True, False, True];
Bit("101") -> to_text() == "101"; .to_numpy() returns a bool ndarray.
[未确认: not exercised against a live database in this pass.]

[chrikrah]

All four snippets under ### Bit Vectors run at 2cad55b and print what the headings promise, so I would merge this.

$ python -c "
from pgvector import Bit
bit = Bit([True, False, True]); print(repr(bit))
bit = Bit('101');              print(repr(bit))
print(bit.to_list())
print(bit.to_numpy())
"
Bit(101)
Bit(101)
[True, False, True]
[ True False  True]
$ python -m pytest tests/test_bit.py -q
17 passed, 2 warnings in 0.04s
# python 3.12.3, numpy 2.5.3, pgvector imported from the PR tree
# not run: the suites that need a live Postgres

non-blocking: a bool NumPy array works too, Bit(np.array([True, False, True])), at pgvector/bit.py:39-52 and tested at tests/test_bit.py:39. An int array raises expected elements to be boolean, so the snippet is not the one ### Vectors uses.

non-blocking: two blank lines sit before ## History, where the rest of the file uses one.

@HUIIIM, was the NumPy constructor left out on purpose?