tl;dr -- an atmospheric entry always loses energy, but sometimes not enough to keep the entering vehicle/capsule/meteor trapped in the atmosphere. It's possible for the entry to change the orbit such that it pops back up above the atmosphere in a lower-energy orbit than before, which might re-intersect the atmosphere, or might still be above escape velocity and never come back.
From experience with spaceflight simulators, specifically Orbiter:
The space shuttle had a tendency to bounce. Everyone says that it is a "flying brick" meaning that it doesn't fly well. Actually it's a great aircraft in its design range... it's just that its design range is mach 10+. The nominal entry profile is to enter the atmosphere at 7000+ m/s horizontally and about -150m/s vertically, with a range to go of about 4000km, with wings level and an angle of attack of about 40deg nose up. If the vehicle held this wings level attitude, it would generate so much lift that it would quickly cancel out the downward speed and start going up. Since the spacecraft is still almost in orbit, it still has almost all of its horizontal speed and the effective acceleration of gravity is quite low and it doesn't take much lift to start flying back up. It literally is almost like a wall at 80km altitude. If this attitude continues, the spacecraft will start flying up, still at almost orbital speed, and arc far over the landing site. I never tried to stretch the glide as far as I could, but based on my experience, I would say that it would probably not enter the atmosphere again until it was thousands of km past the landing site, and bounce again. Each time energy would be lost, so the average orbit altitude (semimajor axis) would continually decrease. Eventually it wouldn't bounce high enough to get back into space and would finish entering, after skipping something like half way around the world.
Since this isn't desired, there are of course tricks to prevent skipping. The technical definition of lift is the portion of aerodynamic forces perpendicular to the direction of motion, just like drag is the portion anti-parallel to the motion. The English word "lift" implies "up", but the math definition does not -- the lift can be directed in any direction perpendicular to the flight path.
The lift could be changed by changing the angle of attack, but there are a lot of tradeoffs between the structure of the wings, the placement of the heat shield tiles, etc which tend to make it preferable to keep the 40deg nose-up angle of attack.
So what the shuttle did, is right after entry, it would bank almost 90deg, so that its lift vector is nearly horizontal. It kept just enough lift in the vertical direction so that the sink rate is controlled. If the ship needed more drag, it would bank more so that it would sink faster and dig into thicker air lower down. Conversely if it needed less drag, it would bank less. As the ship slowed, the total lift decreased (and the effective acceleration of gravity increased) so that it had to roll closer to wings-level to keep enough vertical lift.
Apollo, even though it doesn't have a wing, had an off-axis center of mass. This meant that the capsule would come in at an angle, and effectively the whole bottom of the heat shield would be one big circular wing. It is then controlled the same way, by banking to control the amount of vertical lift. This is called a lifting entry, and all modern manned capsules do something like this. Coming back from the moon, the spacecraft was flying at above escape velocity, so if it lifted back out of the atmosphere, it might go into a very high orbit, perhaps high enough that it would take hours or days to come back. (Considering that by this point the service module had already been separated and the power and life support remaining was only a few minutes, that would have been a Bad Day.) So, the Apollo guidance was designed to control drag as first priority until it got below the local circular orbit speed, at which time it considered itself captured and unable to bounce out. Then it shifted focus to targeting a particular splashdown site.
The alternative for a capsule is to not have any center-of-mass offset. This is referred to as a ballistic entry (as opposed to a lifting entry above). Some of the Mars landers used this, as well as the original Vostok and Venera spherical entry capsules. This design doesn't generate any lift and therefore is never in danger of skipping out.
Similarly, I would expect an ordinary natural meteor to not have any significant lift and not be in danger of skipping out. Meteors sometimes do pass through the upper atmosphere with their original perigee in the atmosphere but above the ground. Sometimes the perigee is high enough that while the meteor burns and glows, it doesn't lose enough speed to drop the perigee to ground level, and therefore heads back out into space in a lower-energy (but still perhaps escaping) orbit. I leave it to you to judge whether you consider this "bouncing off" the atmosphere, but I don't.
Manned capsules tend to use lifting entry, for a variety of reasons. First, it is more controllable and therefore easier/possible to target a precision landing site. Second, because it lifts, it spends more time higher in the atmosphere in thinner air than it would otherwise. This allows it to come in with less acceleration than it would otherwise. The space shuttle coming in from low-earth orbit rarely felt accelerations greater than 1.5g, while a ballistic entry on even the shallowest path from the lowest orbit will hit over 8g.
Soyuz is designed to use either mode -- it normally uses lifting entry, but it can fall back to ballistic entry under certain conditions (usually malfunctions). It isn't fun to enter with over 8g, and it means giving up controllability, but the ability of Soyuz to withstand a ballistic entry has been used four times, and has saved crews that otherwise would have been lost.
