Add ObservationManager - #2575

## Summary

Update the visibility pyramid when a feature gains its first triangulated correspondence or loses its last. Additional correspondences change support but leave the visible feature set unchanged, so repeating every pyramid-level update does unnecessary work.

## Changes

- Put SetPoint and ResetPoint beside the existing zero/one visibility-count transitions, preserving occupied-cell scores for colliding features.
- Extend the existing visibility-score and AddImage tests to cover repeated support, collisions, partial/final removal, underflow and rebuilding the manager.

## Background

The repeated updates are inherited from fe8f82a, "Initial commit" (2016), in src/base/image.cc. The ObservationManager extraction, eaebe3c, "Add ObservationManager (colmap#2575)", retained them.

## Results

Fixed public Add64/Delete32/Delete32 bookkeeping on a 16-image, 64-track, 120-pair fixture:

| Platform | Before | After |
| --- | ---: | ---: |
| Linux/WSL | 1.395 ms | 0.640 ms |
| Windows | 2.776 ms | 1.438 ms |

Each value averages four fresh process means per arm, ABBAABBA, with two warmup and ten measured batches per process. Canonical inputs and outputs match within each platform.

Whole-command Linux observations, including setup/oracles/serialization: median CPU 0.495 -> 0.443 s, median per-process peak RSS 414.50 -> 414.35 MiB, mean instructions 3.879 -> 3.782 billion. Windows median CPU 2.234 -> 2.211 s and median per-process peak working set 14.54 -> 14.33 MiB have overlapping process ranges. These broader counters describe the probe, not isolated pyramid operations.

## Validation

- Current composition: all 18 maintained manager cases and two mapper lifecycle cases passed, without skips. Earlier matched arms passed the same selected counts on Linux and Windows.
- clang-tidy-18 completed on the owner and maintained test with no displayed source diagnostics; generated/header warnings were suppressed. Changed-line formatting passed with the real source filename.
- Incremental build: three translation units and two test links, reusing 46 production objects and 13 archives. Source, selected-link and provider postflight checks completed using the retained outputs.

AI disclosure: Totally vibe coded

## Summary

Update the visibility pyramid when a feature gains its first triangulated correspondence or loses its last. Additional correspondences change support but leave the visible feature set unchanged, so repeating every pyramid-level update does unnecessary work.

## Changes

- Put SetPoint and ResetPoint beside the existing zero/one visibility-count transitions, preserving occupied-cell scores for colliding features.
- Extend the existing visibility-score and AddImage tests to cover repeated support, collisions, partial/final removal, underflow and rebuilding the manager.

## Background

The repeated updates are inherited from fe8f82a, "Initial commit" (2016), in src/base/image.cc. The ObservationManager extraction, eaebe3c, "Add ObservationManager (colmap#2575)", retained them.

## Results

On the composed-fork version, fixed public Add64/Delete32/Delete32 bookkeeping on a 16-image, 64-track, 120-pair fixture:

| Platform | Before | After |
| --- | ---: | ---: |
| Linux/WSL | 1.395 ms | 0.640 ms |
| Windows | 2.776 ms | 1.438 ms |

Each value averages four fresh process means per arm, ABBAABBA, with two warmup and ten measured batches per process. Canonical inputs and outputs match within each platform.

Whole-command Linux observations, including setup/oracles/serialization: median CPU 0.495 -> 0.443 s, median per-process peak RSS 414.50 -> 414.35 MiB, mean instructions 3.879 -> 3.782 billion. Windows median CPU 2.234 -> 2.211 s and median per-process peak working set 14.54 -> 14.33 MiB have overlapping process ranges. These broader counters describe the probe, not isolated pyramid operations.

## Validation

- Clean base and candidate: all 15 maintained manager cases passed in each arm, without skips. The composed-fork version also passed 18 manager and two mapper cases on Linux and Windows.
- Configured clang-tidy-18 checked observation_manager.cc in both arms and the maintained test, with no displayed source diagnostics; generated/header warnings were suppressed. Changed-line formatting passed.
- Incremental qualification reused retained pristine-upstream support and rebuilt the two selected RANSAC-dependent owners needed for the newer headers. Final source, selected-link and runtime-provider checks passed.

AI disclosure: Totally vibe coded

## Summary

Update the visibility pyramid only when an image feature gains its first
triangulated correspondence or loses its last. Extra supporting
correspondences leave occupied cells unchanged, so repeating every
pyramid-level update does unnecessary work.

This is a small inherited optimization, based directly on official
COLMAP 3c1a1b9.

## Changes

- Move SetPoint/ResetPoint into the existing zero/one visibility
transitions.
- Extend the maintained score test with repeated support, distinct
colliding features, partial/final removal and underflow checks.
- Extend AddImage with observation removal and reconstruction of the
manager, avoiding unrelated fork registration behavior.

## Background

The update pattern dates to fe8f82a,
"Initial commit" (2016), in src/base/image.cc. It was retained by
eaebe3c, "Add ObservationManager
(#2575)".

## Results

The source-equivalent transition change was measured in a composed-fork
fixed bookkeeping probe: Add64/Delete32/Delete32 on 16 images, 64 tracks
and 120 pairs.

| Platform | Before | After |
| --- | ---: | ---: |
| Linux/WSL | 1.395 ms | 0.640 ms |
| Windows | 2.776 ms | 1.438 ms |

Each table value is the arithmetic mean of four fresh process means per
arm (ABBAABBA). Each process mean averages ten measured batches after
two warmups. Canonical inputs and outputs agree within each platform.
These are fixed bookkeeping timings, not reconstruction throughput
measurements.

## Validation

- Independent source review accepted the clean-base production change
and adapted lifecycle fixture.
- Clean base and candidate each passed all 15 maintained manager cases,
without skips.
- Configured clang-tidy-18 checked observation_manager.cc in both arms
and the maintained test with no displayed source diagnostics;
generated/header warnings were suppressed.
- Reused retained pristine-upstream support and rebuilt the two selected
consumers of newer RANSAC headers; final source, link and provider
checks passed.
- Separately, the composed-fork version passed 18 manager and two mapper
cases on Linux and Windows.

AI disclosure: Totally vibe coded

## Summary

Integrate the upstream visibility-transition optimization into the fork while retaining its observation-manager behavior and both lifecycle checks.

## Changes

- Merge upstream 137f216, "SfM: Update visibility on feature transitions (CPU perf) (colmap#5006)", as the second parent; retain fork d896761 and its current production implementation.
- Exercise the upstream partial and final observation-removal sequence and the fork registration and deregistration sequence with independently constructed state in the existing AddImage test. Preserve the remaining fork tests.
- Attribute the inherited pattern to fe8f82a, "Initial commit", and eaebe3c, "Add ObservationManager (colmap#2575)".

## Validation

- The selected Linux observation-manager suite passed 18/18 cases with no skips; the AddImage case executed both independent sequences. The changed test and current reconstruction owner were rebuilt, and selected archive members, source dependencies and runtime providers were checked.
- Configured clang-tidy18 on the changed test: no displayed source diagnostics; generated/header warnings suppressed, consistent with retained baseline. Clang-format 18.1.8 left the affected AddImage range unchanged.

AI disclosure: Totally vibe coded