Is there such a thing as a network bridge or is it always meant as a switch implementing a bridging function?
Whilst bridges are described as devices that create separate collision domains in network specifications and tutorials, I have never been able to find any concrete product denominated 'bridge'.
Bridges as a product no longer exist due to how the market evolved. Typically the products specifically called bridges were used in hubbed environments to separate the collision domain or to bridge two network media types together (for example, Ethernet and token ring).
Bridges as a function are definitely still around.
Simply speaking, as a function of separating the collision domain, a switch is a bridge with lots of ports.
For an example of bridging two network media types, access points are bridges that bridge 802.11 traffic to Ethernet.
Yes, a network bridge was originally what we would think of today as a 2-port ethernet switch. They were used to provide separate collision domains without otherwise changing the routing on the network. I always thought the use case was a bit limited, but ethernet switching was the newfangled fanciness as I was coming of age in networking, so the network bridge was already looking a bit long in the tooth as a concept.
The term "switch" is really the term that is nebulous. Early ethernet switches were frequently referred to as multi-port bridges to help define what it was that they were. Later, when the development of routing traffic in hardware some marketing geniuses (tongue-firmly-in-cheek using "genius" there) decided to create the term Layer 3 switching to differentiate hardware based routing from the traditional software based routing processes. This served to further muddle the definition of the term switch, and now we have marketing geniuses (Ibid) referring to "Layer 7 switches" and other such inanities.
As YLearn's answer alludes to, bridges could also bridge between dissimilar media types with greater or lesser amounts of lossage depending on how compatible the media types were...ethernet <-> FDDI was easy, ethernet <-> Token Ring was a bit more challenging to get right.
Today you can make a network bridge in software. Get a Linux box and compile the ethernet bridging support into the kernel. Then you get network devices like br0 which can be connected to interfaces like eth0 and eth1 and will propagate raw, low level packets from one to the other. There is even filter support similar to iptables, as well as a user space tool called ebtables (Ethernet bridging tables) using which you can define rules to filter packets by MACs and other fields or bytes, so your bridge does not have to forward everything.
Ethernet was not switched once upon a time. The original Ethernet consisted of a long cable shared among the stations. Each station tapped into the cable. This is why we still talk abut a "network segment". Today a "network segment" is a logical segment only, but actually looks like a star (or bunch of stars) centered around switches, but once upon a time, it really was like a straight line with stations hanging off it.
If you wanted to connect two such networks, there were two choices: a repeater, or a bridge. A repeater is an analog amplifier device which listens on one wire, buffers the signal electronically and drives the same signal on the other wire (and in the case of networking, vice versa: bidirectionally). A bridge is a computer or embedded system which receives the actual packets, and then sends them to the other side.
Unlike a repeater, a bridge can be smart in that as it passes traffic, it builds a table of known addresses in memory. It knows which station is on which side of the bridge, so it can avoid bridging whenever a station on one side of the bridge addresses a station that lives on the same side. Only broadcast packets have to be bridged, and packets which address across the bridge.
